The SuperReg Power Supply

This will be a post to describe my efforts to more or less regurgitate and resurrect the Walt Jung/Jan Didden power supply design that gained a reputation as a SuperReg.

I want to use this supply for my SG505 project because it deserves and will hopefully get a very quiet and responsive power supply.

A lot has been written about this design over several decades, and many attempts have been made to improve on it, mostly for audio applications. The majority, at least the ones I found, were tailored to hi-end audio power amplifiers, with an obvious focus on supplying lots of power. I need less than 100mA for both rails, which is more in line with pre-amps, DAC's etc. For these type of applications the original design from Walt Jung is more than sufficient. It does not need a lot of special parts, and is deceivingly simple.

The secret to success is to stay close to the original design principles as much as possible, which is what I'm planning to do.

If you want to have some more background on the original Superreg design, the performance measurements and the evolution over time, here is a link with all the information that will keep you busy for a while. SuperRegs  Here is another link to the articles from Walt Jung: directory

I will follow the latest version (2.3) and try to replicated that. The information can be found in the above archive, but this is the document I'm using as the reference for my copy. Version2.3
Here is another refinement/update for the reference: article
If you're interested in buying the original PCB's, you can do that here.

And finally, there is a discussion at diyaudio.com that can be followed here.

Below is a picture of the V2.3 PCB that Jan Didden designed.

I'm going to use SMD components where possible, hopefully without messing-up, and create a version with +17V and -17V outputs and also add the +5V rail I need.

Here is the schematic design I came up with:

Even though I'm hesitant to deviate from the original, there are a few required differences with the original. 

First of all, I added the rectification and reservoir capacitors. This means that I only need to feed two AC transformer winding connections, and leave the transformer, on-off switch, and the EMI filters in another separate enclosure, away from the more sensitive parts.

Because I need to make some changes, I used 499R for R4 and 5K1 for R5 because I'm also not going to use the LM329. It's only available in TO92, and it's pretty expensive. The 4K99 value will cause a current in the Zener of about 2mA. According to the specifications, the device also works with 1mA. However, the recommended combined Zener current by Walt should be around 4mA, so the value for R5/R12 should be 2K7 which sets the current to 3.8mA. The schematic still has the 5K1 value though because I'm not sure yet that I need that much current.

The original information by Walt Jung in the first article about the Superreg also talks about the 499 Ohm for R4 as having the same value as the parallel value of R7 and R6, both 1K, (to better balance the load on the Opamp inputs). The value for the 4K99 most likely comes from a very early version of the supply, called the Sulzer regulator, shown in figure 7a and 7b, in Walt's first document. In that version they use 51K for R5 and R10 (now R12) in those schematics. The voltage setting resistors (R4, R3 and R9 and R8) were 3K16 each. In a newer version, Walt used lower values for these resistors to improve the regulator. My assumption is that he used a factor of 10 to lower the values, and he used wire-wound metal film THT resistors that were only available in the 4K99 value.  

Because I'm going to deviate from the LM329 by using the recommended combination by Walt of two Zener diodes in a back-to-back configuration to keep the noise and the tempco down. It's an old trick to use a normal diode in a back-to-back configuration with Zener, because the tempco's of both cancel each other out. On top of that, a 6V3 Zener is typically at the sweet spot of the tempco for Zener diodes.  I also need an output of +17V. So the voltage output divider values of R6 and R7 have to change. 

In one of the articles from 2/95, there is a table 1 on page 31 that shows different resistor values for different output voltages. There is a 16V and a 18V row, but not a 17V one. In all cases, R1 is 499 Ohm, just as I want to use. I wanted R7 and R14 to stay close to the 1K value to keep the Opamp balance almost the same with the 499R value of R4 and R11, so I selected a 820R value, which is just about in the middle for the recommended 16V (866R) and 18V (806R) outputs.

To get the +/-17V (+/-1%) output voltage with E24 values, I needed to use a trimming resistor, so R6a has a companion of R6b, the same as with R13a/b. 

The formula to calculate the resistor values is :  Vout = Vref x (1+ (R4/R3))

This means that we need a 1K25 value for R6 and R13 with the 6V9 reference (although that value needs to be confirmed), see actual measured values below.

The parallel combination of R6 and R7 with these values is 495 Ohm which dictates R4 to be 495 Ohm as well. Keeping the 499 Ohm value should be good enough initially, but can be changed later on.

The recommended Opamp is the AD825, Jan Didden mentioned that the NE5534 can be used, or even has to be used for higher supply output voltages anyway, and since I already use them in the SG505, I might as well use them here too. They are a lot less expensive at 1/10th the price of the AD825. You could swap them out if you want as long as you use an SOIC package. I also ordered two AD825's which I'm going to use first.

Most of the parts are SMD, with the exception of the larger capacity electrolytes, and the 10nF film capacitors for the rectifying bridges.

It seems that Walt very carefully selected the series transistors, the D44H11/D45H11 so I'm going to use them too. They are not expensive.

The bias transistors for the series transistors I'm using are the equivalent SMD versions of the BC546/556 TO-92 version.

Because the required output is +/-17V, I raised the voltage for the series Zener diodes (D2 and D7) to 7V5, to keep the Opamp output just about centered within the rail voltage, as is recommended. The Zener values are not overly critical, I just happen to have this value in stock.

Note that the recommended Panasonic HFQ series for the 120uF/35V electrolytes have been discontinued for a very long time, so I selected decent quality replacements.

I will need small heatsinks, so I will not put a footprint on the PCB, just use the U-shape parts that you can screw on the TO-220 package.

I've also added the +5V section, and tried to keep that out of the way of the other two supplies as much as possible. That's why I tap the AC inputs from both transformer windings, the keep them balanced and separate the two supplies at the AC level, and also create a "digital" GND for the Arduino Nano and the relays I will use with the next version of the generator.

The layout is now done, and the request for sponsoring from PCBway for the production and shipment has been done. They generously support my activities so I can spend the money on parts.

Don't worry about the apparently reversed picture of the terminal blocks, they will be installed correctly. (;-))

The parts and the PCB arrived and I quickly put together a working version, again using my new reflow hotplate.

When I first powered it up and checked the output rails, I was pleasantly surprised with a +/-18.9V output. This is without the trimmer resistors. 

However, when I tested the +5V supply I was greeted with a 0V output. Bummer! On careful inspection, I spotted my mistake. I was planning all along to tie all the common grounds to the GND pin of the output BNC. When I was testing the supply, the digital GND for the +5V was still floating and not connected to either common. Well that makes sense. However, when I thought about it some more, I realized that I did not need to connect the digital or DGND to the analog common GND's. I could keep the digital supply completely separate because it will only power the Arduino Nano with the OLED display, and the relays I'm planning to use for the range switching. I could keep all of that "out of the way" from the generator by creating a dedicated and separate DGND. That required a small modification of two diodes and two 10nF capacitors to tap the other two windings and create a negative return path.

While making the changes to the schematic, we now have a V1.1:

Note that the two Opamps are still shown as the NE5534, but I installed the AD825. (KiCad does not have a symbol for it so I'm leaving the NE5534 as a place holder)

The two optional diodes and two filter caps for the separate +5V supply DGND are added underneath the board.

First step now is to tune the voltage rails.

I first measured the Zener diode combination. The plus supply was 6V99 and the minus supply was 7V01. They are very close to each other. When I used a 6K2 trimming resistor for R6b1 and R13b1, the voltages after warming up are now +17V04 and -17V04. Excellent!

The parallel resistance of the 1K3 and the 6K2 are now 1K075, very close to the original 1K. 1K075 in parallel to 820R makes 466 Ohm, I could lower the 499 to 470 Ohm to better balance the Opamp inputs but I'm going to leave the 499 Ohm as is for the time being.

The downside of using hi-efficiency SMD LED's for the current source is that they are very, very bright. I think that I'm going to use a black pen to darken them a bit.

I used my thermal IR camera, and was surprised to see that the current source transistor (Q2) was getting a lot warmer than I anticipated. I am using the SMD version instead of the original TO-92, but even so. There is 1.97V across R1, so with 240 Ohm, that's 8mA of current. The earlier version (Fig 8a and 8b) used a 100 Ohm value that Jan Didden changed to a 249 value. I used a standard 240 Ohm value.

To the right is the positive supply with the (hot) 330 Ohm load, to the left is the negative supply with no load. The green LED's are also contributing to the hot spots on the display.

Even without a load, the voltage differential with an input voltage of 36V across the series transistor makes them warmer than I had anticipated. (you can easily see the difference between the left unloaded and right loaded versions)

With a 330 Ohm 1W resistor as a 50mA load, the series transistor gets a little warm but not overly so. Because the generator setup I have now will only need about half of that current, this will be OK. If not, I can use a larger heatsink.

Next step is to make an FFT of the outputs to see how quiet they hopefully are.

This is the positive supply, also with a 330 Ohm/50mA load. Can you believe that I checked and double checked my measurements? This is unbelievable, wow! There is no difference with a shorted input lead, so the supply does not add anything it seems.

With the 50mA load, the input to the regulator has a 1Vp-p mains ripple, that's completely gone, and no harmonics or noise. Too good to be true? 

I was planning to use a mains input filter, I'm reconsidering that for the moment. Even in my noisy environment, there is nothing to worry about.

Here is the negative supply:

Walt, I'm mightily impressed, hats off for your design.

I was already formulating plans to modify the +/-17V supplies to a shunt version that Walt published in a later document, to separate the raw DC side even more, but that is not needed in my case.

And now the final test, the generator with the new supply:

A very clean output with only a tiny bit of mains related harmonics on the left.

I do not really trust the measured and reported THD+N number of 0.0056%, but visually, it looks great with only a tiny third harmonic spike visible, and this is still without any shielding whatsoever.

This power supply is great and that part of my problem is solved!

The +5 Volt circuit

I'm going to hi-jack this SuperReg post by showing information about the +5V circuit I added. It has nothing to do with the SuperReg, but I'm adding that adventure here because I put it on the same PCB.

Here is the FFT from the loaded +5V supply with the positive and negative still loaded with the 50mA:

This is with a 51R resistor for 100mA load. The 50Hz mains harmonics comes through but is otherwise very clean as well. Even when I extend the measurement to 100KHz, there is an uptick in the noise floor after 30KHz, but no additional hash. 

But...

When I connect the Arduino counter circuit to the +5V supply, I get this...

My solution to create a 5V rail is causing an issue, and with it, the transformer gets very hot so something is very wrong with my design. The diodes must be shortening the two transformer windings or something like that.

The Arduino only draws 14mA, that's not the problem. I need to figure that out.

I now modified the 5V input by tapping only the transformer winding for the positive section, by using two diodes. Of course, the three power supplies still get GND connected on the generator PCB, because I need to tap the sinewave as an input to the counter. Now that I don't connect the two windings of the transformer, there is no current flowing between them anymore, and the transformer stays cool. That part of the problem is solved, but I'm not there yet.

This is the disappointing result:

A bit better, but no success whatsoever. There are two possible solutions I can think of. One is to galvanically separate the pick-up for the counter from the generator, by using a pretty fast opto-coupler. The other possible solution is to use a separate transformer for the 5V supply. That's the easiest solution to try.

Here I'm using a separate 12VDC Wallwart just to give it a quick shot. I also tried my Lab Supply with an 8VDC input, with the same results.

A lot better, but not good enough yet. It proves to me that I need to galvanically separate the output amplifier of the generator going to the Arduino counter.

Here is also evidence why I don't trust the THD+N results from the software. You can clearly see more harmonics, but with a better result than the earlier 0.0056% ???

I'm now using an H11L1M opto-isolator, that has an internal amplifier and Schmitt-trigger gate. In order to get enough signal drive for the transmitter, I had to bridge out the output capacitor that you can see in the amplifier circuit above. That works, so the counter is functioning again. However, there are still added harmonics. The solution I think is to use an additional output amplifier buffer, solely for the counter input. As long as it produces a logic level signal, the opto-coupler will be happy.

Here is the result with a hodge-podge of wires and an additional protoboard to test the opto-coupler:

Getting better, but I'm not happy yet. If it turns out that driving the transmitter is causing the added harmonics, I will need to reconsider if I want to continue with the counter. (PS this time the THD+N is more realistic I think. Go figure...)

As a next step, I created a simple sine wave to square wave amplifier and let that drive the opto-coupler to create more isolation with the generator. Now we're finally getting there:

Here is the amplifier section with the circuit to connect to the Arduino Nano:

I'm done with this post for now, and will continue with the SG505 post.

I have purchased another transformer for the SuperReg that I will try later on. It has a copper shield around the windings that can be connected to Earth GND for extra shielding:

Update to a new version

I like this SuperReg very much but because I made a mess with the +5V supply, I decided to create a new version 1.1 that removes that circuit and only has the positive and negative supplies on it. I will create the layout such that these two sections can be easily separated so you can use them individually, and use one extra side to create another supply. To apply the lessons learned, make it more universal and also support higher currents, I will allow space for larger electrolyte smoothing caps for the raw DC supply, create a bit of a PCB heat-sink for the current source, create room for larger heatsinks for the series transistors and provide a ground plane while keeping the "star" GND connections.

Visit often and stay tuned, there may be more...

DIY Build of the Tek SG505 Oscillator

 

The rather famous SG505 oscillator

I'm going out on a limb here. Although I already built a copy of the SG502, described here, the SG505 is a very different beast all together. At the time I did the SG502, I was not ready to attempt the SG505, a masterpiece of analog engineering by wizard designer Bruce Hofer. After a number of years mulling about it, I was recently looking at the instrument again after seeing it used in a YouTube series, and this time decided to give it a try.

As with many, if not most Tektronix designs, they incorporate special parts, or the parts are obsolete by now. The added difficulty is that with these older designs (this one was created in 1980) that pre-date the internet, you can't find any datasheets for these parts. Most manufacturers went defunct or merged and do not show these old parts on their websites.

So when looking at the SG505, I first wanted to try to build the sine wave oscillator. I used the information from the Option 2, that has the balanced outputs, because that has improved schematics and parts.

Here is where you can find a lot of information about the SG505 and it's designer Bruce Hofer.

So what is so special about the SG505

First of all, among specialists, it has the status of a real piece of analog art, designed by one of the great masters himself, Bruce Hofer. While at Tektronix, Bruce was responsible for many products and designs in the 7000 oscilloscopes and the TM500/5000 series. He also designed the AA501, the first automatic audio analyzer that worked in tandem with the SG505. As a matter of fact, I understand they designed the AA501 to verify the SG505 after production. In the manual for the SG505, you can see the rather elaborate method to calculate the total harmonic distortion, and the equipment needed. 

Just the specification of 0.0003% THD from 20Hz to 20KHz will give you an idea of the design quality. 

The SG505 has at least one patent that I know of, describing a "State variable oscillator having improved rejection of leveler-induced distortion". Have a look here : patent.

When Tektronix management was not interested in taking the AA501 and later the AA5001 distortion analyzer concept further, Bruce co-founded Audio Precision and that is the current and undisputed leader of this field.

You may ask why I'm so interested in rebuilding the SG505? First of all, it's a great design made by one of the few analog wizards. It is very well documented and also has been manufactured for several years so the selection of components went through this test. This "fitness for manufacturing" should not be underestimated, because the parts (source, value, tolerance, tempco, etc) in every possible permutation will have to meet the specification of the final product over the years. In other words, the bugs are out. This cannot be said about many other DIY designs.

The other reason for selecting the SG505 is that you can freely select any frequency by just selecting the desired range, and using the dial and vernier adjustment to select the frequency you want. There are several other pure oscillator designs that are more modern, and even have better specifications. In almost all of their applications, they are used in a THD measurement, and in those cases, you "only" need a few fixed frequencies. That makes it easier to get better specifications because you can tune both sine and cosine oscillators with fixed parts. None of them use an analog variable frequency selection.

Some Small print:
As I normally do on my Blogs, I describe the way I get to the final design, with trials and errors, ups and downs, warts and all, but hopefully in a way that you learn something, and avoid my mistakes. With my projects, it's not going to be an IKEA hand-holding type design. Some effort is required. And most importantly, there are no guarantees it will work for you too, I'll do my utmost but that's about it.

I will add information to this Blog and the Github as I go, and with this pretty extensive project, it may take a long time. As a matter of fact, I may not be able to finish it with the sought after quality, so take that into account before you heat-up your soldering iron or start to order parts. I'm not a really experienced engineer, and I'm not hampered in any way with specific know-how or experience, nor do I have the proper equipment to verify and test the design. I'll do my best, and will make you part of the process, but that is as far as I can promise.

Interesting articles and information:

There are many articles about sine wave oscillators, because building a simple one is easy, but a really good one is a real challenge. Over the years I have collected a few interesting ones:

Rod Elliot has a lot of interesting information on his site, but there is one particular posting where he explains the various architectures with their pro's and con's.
https://sound-au.com/articles/sinewave.htm

A few other articles can be found on my Github site here:
https://github.com/paulvee/DIY-Rebuild-of-the-SG505.

Then there are several Forum posts that deal with the modernization and rebuild of the Bob Cordell THD Analyzer that contains a very good oscillator, just like the SG505 (that original article is on my Github). The latest and most impressive rebuild of that instrument that is 30 years old, but now with a processor to drive the measurements is on this Github from Emile666, a very impressive and complete design. Also have a look at the diyAudio site, there are designs that deal with specific 1KHz and 10KHz pure oscillators of outstanding specifications (search for "LT AN67 1KHz or 10KHz oscillator", from Frex) The AN67 application note is listed on my Github. The author of that impressive design is the rather flamboyant Dale Eagar, who worked for my buddy in crime Bud at LT. Small world.

Component Issues & Challenges

For this particular project, let's discuss the road-blocks and challenges for a rebuild, based on the components that are used in the instrument.

First off, there are several special parts and some of them are no longer available on the market.

The potentiometer that sets the frequency is a high quality single turn dual 10K wirewound version without a stop. It was originally made by Spectrol, but is now part of Vishay. The part used is a Model 100 with the following part number: 100B2-103-103-XXXX

According to the datasheet, it has +/- 3% tolerance and the independent linearity between the two sections is only +/- 0.5%. Unfortunately, the datasheet does not specify the THD. This part is still active and manufactured it seems, but nobody has it in stock. 

The next item is a mechanical part that is used in combination with the potentiometer, to create a finer adjustment by reducing the number of rotations 6:1. The Tek part number is 401-0161-00 and the manufacturer is Jackson Bros (London). Their part number is listed as 4511/DAF in the SG505 manual, and believe it or not, you can still buy this part for a reasonable price, here. This reduction unit also gives a very smooth and controlled action on the potmeter. Highly recommended!

I am the lucky owner of having the potmeter and the reduction unit, it is used in my SG502, but for others, I'm going to try to find a replacement part or method for it. It's the most critical part of the design so it would be critical to find an alternative. I have some ideas using other potmeters, even digital ones but I have no experience with them whatsoever so I need to give that a try when I have a functioning oscillator.

The next item in the list is the special drum switch (or cam shaft) that Tek manufactured themselves that select the output attenuation. This in addition to the special pushbuttons that are used to select the frequency multipliers. In my design for other Tek instruments, like my 5CT1 (without N below) where I added readout capabilities and also for the SG502, I used reed switches or a small relay in combination with a simple rotary switch and diode matrixes to re-create the sometimes rather complex switching arrangements. 

Luckily, the SG505 switching used is quite simple. At first I was contemplating using analog CMOS switches or MOSFET's, but I understand that will add too much resistance or add to the THD distortion budget, so I will most likely use reed switches again.

Next-up is the J-FET used in the AGC circuit. The Tek part number is 151-1021-00 and is used in many Tek products. In the manual, they show the FN815 from Siliconix. I have not been able to find any information about that part at all, and I know I'm in good company because of the frequent use in other products. Many makers are looking for replacements. There is another J-FET 151-1025-00, an SFB8129 (in later manuals changed to SPF3036. The recommended alternative is the 2N4416, and the alternative for that one is the J111) This J-FET is used in the power supply as a current source, but that one is a lot less critical. Lastly, that same J-FET is also used in the output buffer amplifier in a special circuit that Bruce got a patent for : patent

One of the most critical components, in terms of the distortion budget, is the special hybrid substrate that is used to select the dBm output settings. I happen to know that the construction used in the SG502 is the source of some extra distortion, so Bruce Hofer created an unofficial modification to change that for the SG502, significantly reducing the total harmonics from an original 0.035% THD to well under 0.002%. My assumption is that using discrete components is going to be a challenge, but on the other hand, we now have very good SMT devices Bruce didn't have at the time.

Lastly in this list is the set of matched timing capacitors used in the oscillator phase-shift amplifiers. Luckily, for a DIY project, the absolute values that are matched in the SG505, are less critical. In a manufacturing process, you don't want to tune the product, it has to work right of the production line with as few adjustments as possible. The critical factor for the original design was to create a situation by which the frequency dial is set to say 1, and by selecting the X10, X100, X1K and X10K, you would would get the exact reading of 10Hz, 100Hz, 1KHz, 10KHz and 100KHz +/- a few percent, but we don't really care do we? Besides, we don't have the frequency dial anyway. We will need to work around that.

I already have a little and simple side-project working, using an Arduino Nano and a small OLED display to measure and show the frequency.

Simplifications for the DIY design

From the outset, I will be concentrating on the sine wave oscillator itself. I'm not planning to add the differential output that the Option 2 provides. What I will eventually use as the sync output is still open. All I want is a way to trigger my DSO on the signal, and an edge of a square wave is actually more stable to trigger on than the slope of a sine wave. 

I will also not implement the Intermodulation Module because I don't see the need for it in my projects.

Next up is a more simple instrument setup. The original SG505 has to function in a TM500 or TM5000 mainframe, together with a host of other instruments right next to it. Some of them can be very unruly and noisy neighbors. The SG505 is completely floating from real earth ground inside this environment. My plan is to keep it that way, but build it in an enclosure that has no earth ground connection anyway, so it will be floating from the outset.

The power supply is a critical element in the chain and will deserve a lot of attention, but I will address that later. I will use my lab dual power supply in the beginning.

My Plan

My plan is to build a prototype PCB that will give me a functioning oscillator that will allow me to test and try some different strategies and components, and build a reference platform for my measurements using my mostly home-brew set of inexpensive tools. Some if which I already have, some I may need to build to verify and test the specifications. I envision something similar as the design of the VBA Curve Tracer, where we went a step at a time over the course of about two years.

After going through the documentation many times, and watching YouTube videos and strolling the internet for information, I started by creating a schematic in KiCad. I then spent many long hours pouring over the Replaceable Electrical Parts list selecting and finding the parts for the prototype. I used Google, Mouser, DigiKey and LCSC for days, and eventually decided on a BOM to order from LCSC because of the total price including S&H. It turned out that they did not have one particular resistor value from all the parts I need, so I ordered a combination of two to get the missing part.

I did not save money on the parts but selected the best quality. All resistors are 1% metal film, better than the original THT resistor specifications. I also matched or exceeded the higher precision resistors (like 0.1% at 25ppm). However, I did not use Mica capacitors, but good quality NPO with the same 1 or 2%  tolerance. The electrolytes are all Tantalum except for the two used in the supply rails.

Almost in parallel, and now knowing the parts and footprints, I started to design a PCB and ordered it as well. Both the parts and the PCB's will be on their way soon. Talking about on their way, another "way", PCBWay has agreed again to sponsor me with this project, and I'm grateful to get boards from them for free and with a very high quality. I may need to go through a number of re-designs while progressing so this support is very helpful.

The schematic for the prototype

There are a number of rather unusual refinements that Samuel Groner pointed out in a diyAudio.com post. Here is that post replying to a user that wanted to improve the SG505:

Bruce Hofer did (as usual) an excellent job here. There are a lot of subtleties going on (and I'm sure I'm missing one or the other), and it is unlikely that we'll find an easy fix which will significantly improve performance. Just to highlight a few details:

* It looks like residual ripple of the peak level detector is cancelled through C1502 and C1603;
* The NE5534s implement various tricks--feed forward compensation (C1514, C1622), two-pole compensation (C1405, C1406, R1405), forced class A operation (R1621) and even a simple composite opamp (Q1620, see US-patent 4296381); [My comment: this is in the output amplifier section described quite a bit later here]
* The for higher frequency ranges unused integrator caps in the second integrator (C1312, C1320, C1322) are pre-charged to a specific voltage through R1421/R1422. I presume this speeds up settling after a range switch, because the capacitors are already in the correct phase relation to those in the first integrator (which a pre-charged to 0 V);
* The power supply of the peak detector is heavily low-pass filtered (R1611, C1600). This localizes the current loop of the distorted currents, and avoids coupling to the main signal path.

One thing you could try is to make the drain feedback of Q1501 adjustable. Unsolder the ends of R1510/R1513 which connect to the gate of Q1501; take a (quality multi-turn) 1k trim pot; connect its wiper to the gate of Q1501 and the resistance element ends to R1510 and R1513 respectively. By adjusting the trimmer you might be able to reduce the 2nd harmonic for certain frequencies. But make sure it doesn't get worse for others!

Something else might be the addition of a 68r resistor between pin 5 and 6 of U1401 (as done for U1520). This forces the output stage of the NE5534 into class A operation, with possibly some distortion improvement.

This is after I fixed the discovered errors.

The component names and values are the same as in the original schematic in the Option 2 manual. Where they do not, or I had to use other parts, I indicated that.

However, I deviated quite a bit from simply redrawing the original schematic in the manual, to show the many intricate and not-so-obvious feed-back and feed-forward connections, as well as the many compensation circuits, especially around the inverting Opamp U1510. This is where most of the magic happens.  I suggest you read the Theory of Operation in the manual and also the patent application for the feed-forward invention, that allowed Bruce to significantly reduce the harmonics. In essence, he applied his wizardry to remove the fundamental frequency AC signal from the AGC circuit DC-level in several places. I know that the picture above is not of great quality, but a Github project is available here and I will upload files as the project progresses.

The three hierarchical boxes in the schematic are "hiding" the frequency and dampening switches and parts but are inconsequential at this moment. I just followed the original design, but used jumpers instead of switches.

The 2-layer PCB prototype

The layout of the PCB largely follows the schematic, so there is a full circle with the circuits following the State Variable design as is in the schematic as well.

I made the decision, right or wrong, to use ground fills on both sides of the PCB. We'll see if that will show up as an issue, in which case I need to go to a "star" grounding method with all ground going to the output BNC connector.

For this prototype, I will be using jumpers to select, add and try components and also activate the capacitors for the frequency selection and the AGC damping. For the frequency selection, I will use a normal dual (stereo) cermet 20K potmeter, just to have a functioning oscillator. 

J-Fet Selection

I added two kind of foot prints for the J-FET so I can use SMD and TFT to experiment. The TFT version has a socket. I actually have the original but now elusive unobtanium J-FET, and I also have a Curve Tracer, described here, so I can verify or measure the specifications if needed. I already ordered or had a small selection of J-FET's to try.

The elusive, unspecified & unobtanium 151-1021-00 J-FET

While waiting for my parts, that were returned to sender because a very bright DHL delivery person could not find the apartment I lived in for 17 years, I started to look for alternatives for the 151-1021-00 JFET. I already had a number of candidates, and more are in the shipment so I could get started. As a spoiler alert, I think I have been successful, but I now need to test the alternative 2N4391 in the real circuit, and compare the results with the original 151-1021-00 part.

Because I did not do a write-up for testing JFET's with our Curve Tracer yet, I added that episode to the following document. Scroll down to the end in that posting to see the JFET result.

Making measurements with the VBA Curve Tracer

Capacitor Selection

For the frequency setting capacitors, I used good quality Polypropylene film capacitors, but now that modern NPO and COG ceramic capacitors are available in good quality and with very precise tolerances, I may want to experiment with them as well. I will most likely trim the Polypropylene values to match by using the ceramic parts, and measure the effects.

What are the chances of success for the rebuild?

I'm very anxious to see what the results of this prototype is going to be. When I built the SG502, I was pleasantly surprised that my rats-nest construction (no PCB) produced a slightly better THD specification than the production version one had. A testament to the conservative specs and the quality of the original design. 

Who knows what I find with this one...

How am I going to verify the results?

As I stated earlier, I do not possess all the proper equipment that is required to really verify the SG505.

I will have to do with less and see how far I can get. If needed and possible, I can build something, like the passive notch filter (067-0938-00) that the Tek engineers designed specifically to test the SG505 performance. 

I already have an active T-notch filter for 1KHz that I may need to duplicate or extend for the other frequency ranges (10Hz, 100Hz, 10KHz and 100KHz), we'll see about that when we get there.

Luckily, I have a kind-off reference oscillator, the Viktor Mickevich 1KHz design that has a very respectable THD. He claims (verified by others), a whopping <0.00001% THD. 

With my simple DIY equipment, (look at this post and this one) I could measure the THD for his oscillator as far down as 0.00096%, so that is a very good start.

For the rest, time will tell.

Possible solutions as an alternative to the dual 10K potmeter

As I mentioned earlier above, this frequency adjustment potmeter is a very critical part that is crucial for the operation and more importantly, the specification of the oscillator. The specification needs to be tight because of the "State variable" design of the oscillator. Each section of the pot sets the frequency for the two 90 degree phase shift circuits (states). If there is a difference in the resistance between the two potmeter sections, it will negatively impact the phase shift. The timing capacitors also have an equally important role, but they can be more easily matched to each other or the value tuned by adding parallel capacitors.

The combination of the original 1-turn 360 degree 10K potmeter together with the 4511/DAF 6:1 reduction allows the sensitivity or rather resolution to be as follows. For the potmeter, 10K / 360 degrees is 27.7Ohm per degree of rotation, further reduced by 6 to 4.6 Ohm per degree of rotation. And then there is the vernier adjustment of +/- 1% of the frequency. 

Looking at it from a frequency perspective, with the frequency multiplier at 1K, the potmeter can adjust the frequency from just below 1KHz to just over 10KHz. That translates to 1,000 Ohm for every 1,000 Hz, or 1 Ohm per Hz. 1 degree of rotation is 4.6 Ohm and therefore 4.6 Hz. The vernier range is 1% of the frequency so that translates to 10Hz at 1KHz and 100Hz at 10KHz. That will be not so easy to duplicate.

So, provided you can't get your hands on the original Model 100 potmeter, what are the possible options or alternatives?

After quite some time looking at alternatives, I came up with a few potentials.

1. Use a dual (stereo) cermet potmeter
With the right 6mm shaft diameter, it can also be used in combination with the 4511/DAF 1:6 mechanical reduction. The challenge is that these potmeters do not have very tight specifications. A good quality one will still cost about 25 Euro's and at best has a 10% specification, but I could not find a specification for the linearity between the two halves, and that is the more critical specification. An overall value miss can be adjusted with resistors at the beginning or end, but not in-between. This type potmeter is what I'm going to use with the prototype. I can use a trimmer to make sure both halves have the same value at a particular setting.

2. Use a dual 10 turn precision potmeter
This will give you a much higher resolution in setting the frequency, and you don't need the 4511/DAF reduction. The vernier circuit can stay as it is. The specifications for tolerance (3%) and linearity are better than for a cermet potmeter, but there is a significant price differential. A suitable one will cost about 46 Euro's. The downside is that you need to turn the knob a lot (10x) to go from the beginning of a range to the end. Even with this quirk it could still be the most viable solution. I searched at Mouser for candidates, and by the time you have made the criteria selection, there are only 4 left, ranging in price from 46 all the way up to 157 Euro's. 

The least expensive one is the Bourns model 3549S-2AA-103/103A

It has a linearity specification of 0.2%, an overall 3% tolerance and 50ppm/C. This is pretty close to the original Model 100 that has 0.5%, 3% and 20ppm/C.

3. Use a digital potmeter.
I have no experience with using them, but I first thought this could be the solution. So after reading a lot about these devices, I found that there are actually a few stumbling blocks for this particular application. With most devices, the voltage range of the potmeter is limited to 0..5V. The regular maximum granularity is 256 steps (there is a special one with 1024 steps). This means that from a 10K value, every step will be 39 Ohm. That's 8.4 times more than the original resolution, so the vernier adjustment has to be modified to cover the difference. That can be done, but digital pots have other issues, like the linearity and tolerance, and then there are the resistance of the wiper and, a bit less important, the end and beginning resistance. When you change from one setting to the next, there could be a jump or glitch in the frequency, although there are ways (zero crossing) to avoid or suppress that. On top of that they are not that good with temperature changes. 

Based on our particular circuit, one of the best solutions could be the TI PGA2310 designed for hi-end audio stereo volume control. It has excellent specifications, even removing some of the issues with other digital pots. The chip is not cheap however, with a price of  about 28 Euro's.

Most importantly, these digital pots will need some form of digital logic to drive them. Normally, you would use a rotary encoder as the input, and use logic or a processor to drive the electronic potmeter settings. I have not discounted this solution, it provides some potential but it's not my favorite to develop at the moment.

4. Use a rotary switch.
When using a rotary switch, you need one with a double deck and one pole per deck with at least 10 positions, so you can create 10 segments of 1KOhm each by using fixed resistors. By using tight specifications for the resistors, you can really get the best linearity and matching between the two sections and probably also get the best Tempco results, surpassing the original design. The downside is that you can't smoothly turn through the whole frequency range, you have to select a section (1/10th of the range) and than use the modified vernier (modified to 10x the adjustment) for a finer resolution within the section. You can select shorting and non-shorting versions for these switches to avoid large frequency jumps when you switch. 

The suitable devices I found will set you back by about 45 Euro's. This is a rather complicated solution for a rather cumbersome user experience, so not my favorite to develop. For almost the same 45 Euro's, you could also buy the dual 10K Bourns.

Summary:

Depending on your particular application for the oscillator, there are a few options you can select from.

Frequency Display

As mentioned above, because we don't have, or will have a dial, I'm planning to use a small OLED display to show the frequency of the SG505.

Here is the prototype with an Arduino Nano and the tiny display I'm planning to use. I like the yellow color. The dimensions are 0.9" with 128x32 pixels.  I tested the FreqCount library from Paul Stoffregen with my function generator, and it easily covers the range of 10Hz to 200KHz.

The boards arrived.

I received the boards from PCBWay within a week of posting the Gerber files. They are really fast.

There is no gold plating, that is the light fixture playing tricks. The quality of the boards is great, as I came to expect from PCBWay. They also sponsor me for this project which is highly appreciated!

Update on the parts ordering saga:

Unfortunately, the parts I ordered at LCSC are still in limbo. They have been returned by DHL to the European distribution center and will then be sent out again. Hopefully this time DHL selects a more intelligent driver that can find the apartment I have lived in for 17 years. In his ultimate wisdom, he determined that I don't live here, and instructed to have the shipment returned to sender. DHL was unable to reverse this mindless decision. Great! The order was placed on March 4 and I'm told it may take another 1-2 weeks before I can get my hands on them. I just received notice (April 2nd) that the parcel has been re-discovered and this time I instructed DHL to deliver it at a pick-up point so I can pick it up tomorrow. Finally.

Well, on April 3rd, I finally have the parts in my hands and can continue with the project.

To create the best possible frequency selection, I selected the best matching capacitors for the timing ones. Because my minimum ordering quantity was 5 for the 1uF and 20 for the 100nF, 10nF and 1nF all with 5% tolerance, I could easily get matching pairs, but not precise values. That can be tweaked later if needed, but I don't think I need that. The matching between the values is much more critical for the distortion budget.

I also should mention that I'm using the Tektronix 151-1021-00 JFET in the AGC circuit. I'll investigate alternatives later.

First power-up and test

I used my new hotplate reflow solder station to solder all the SMD parts, and that went very well. I then added all the THT parts, and after thoroughly soaking and cleaning the board twice and inspection with a microscope, I tested the power rails for shorts. No issues, so I then used my Lab supply to provide +/17V at 50mA each and gave it a go. My DSO was showing a nice sinewave! Great!

While going through the range selection with the jumpers, I noticed that I goofed with the schematic and subsequent layout for the 10K multiplier selection jumper J15. A few trace cuts and jumper leads fixed that. I also noticed that turning the frequency potmeter didn't really change the frequency. That was caused by me flipping the 3-pin connector the other way around for one half of the potmeter on the PCB. Swapping the connector leads also fixed that. I now had a fully functional sine wave oscillator.

Frequency selection

Because I use a "kind of" dual 10K potmeter (see above), but with separate adjustments for either half, my plan was to see the effect of "in-equality" of the two pots, and adjust them to be exactly the same by using an ohm meter. The goal was to see what effect the in-equality (not only the linearity) would be on the distortion. I will get to that later. However, because of my goof with the connector swap, I stumbled on an optical verification of the in-equality or tracking. When you swap the connector, so one half of the pot will drive one phase change circuit with an increasing value, and the other phase change circuit with an "equal" decreasing value, the frequency should stay at the same value. But only if the two pot halves track each other 100%. Mine does not, there are slight changes in the frequency when I turn both pots together. 

Here is what I'm seeing. At fully CW or fully CCW, the frequency is almost the same at say 308Hz. During the rotation of the potmeters together, I see the frequency dipping down to a minimum frequency of 259Hz somewhere at 75% of rotation and then rather rapidly going up to 308Hz again. My calculator shows that to be an error of a whopping 19% in frequency which is quite a lot. 

To put that in perspective, the 10K potvalue creates a frequency range between 83 and 1.100Hz, or 1.017Hz. That means 9.83 Ohm/Hz, so a tracking error of 308-259=49Hz/9.86Ohm=5.96% of 10K. That's more reasonable, but it shows how important the tracking specification is. And we don't even know yet what the distortion contribution of this tracking error is.

I will do that same measurement with my original Model 100 precision potmeter later.

Does it meet spec.?

I use my dual lab supply to supply +/17.00V. The current consumption is 24mA for the positive rail and 26mA for the negative rail.

The sine wave output is 2.04VRMS (6.00Vp-p) into 1MOhm , and that is according to the specification.

The frequency ranges are within specifications.

• 10x     8Hz to 111Hz
• 100x   83Hz to 1.1KHz
• 1Kx    833Hz to 11.7KHz
• 10Kx  8.8KHz to 107KHz

When I go beyond 107 KHz, the sine wave collapses, so that's really the maximum.

The Vernier adjustment at a frequency of 10KHz goes from 9.92 to 10.12KHz, also good enough.

The AGC circuit works well, visually, but I did not test that any further.

When I set the frequency in the 10x multiplier setting, and adjust the frequency to mid-range and then select the other multiplier settings I see this:

• 10x     50Hz
• 100x   515Hz
• 1Kx    5.26KHz
• 10Kx  55.2KHz

This to me a fine, but can be further tweaked by using trimmer capacitors.

So what is the distortion?

I had some issues to find and load the drivers for my modified Creative EMU-0202 USB "Sound Card" (look at this post for more information), after I went to a new Laptop a while ago after a major W10 induced debacle. Most of the installed software had to be installed again, but I didn't need the FFT capabilities until now.

After some experimentation, here are the results with the PCB just lying on my desk. This is with the range setting at  x100 (100-1000Hz), and using jumpers instead of the frequency select potmeter, to eliminate the leads and the potmeter tracking differences. That puts the frequency at the top of the range at about 1.1KHz. I'm using a DIY 600 Ohm -10dB attenuator, and the E-MU0202 and the ARTA software is calibrated with it. Note that this is taken straight from the oscillator output, and not from a "real" output amplifier. I do not yet know what impact the 600 Ohm loading is for the circuit.

Not a bad result I think, but not near the specification of <0.0008% in the 20Hz-20KHz range.

After several hours trying, searching, testing and pulling some valuable hair, I now know why these results are too far away from the original specification. First of all the settings in the ARTA software.

Second, the 600 Ohm -10dB pi attenuator I was using to feed the E-MU with a lower input signal is indeed loading the oscillator circuit and is the cause for the higher THD distortion. Rather than showing all the previous results I made, I'm now reverting to the setup with the 100K loading of a potmeter at the output of the oscillator circuit so I can reduce the signal amplitude. 

This is the result:

The specification is <0.0008% THD so we're significantly below that figure now.

As a reference, I also made a measurement using the Victor Mickevich oscillator, and that shows this:

It seems my measurement setup is OK, apart from the larger amount of noise. Earlier measurements from a few years ago using the same modified E-MU0202 and oscillator, but with older versions of the software and drivers and a different laptop, produced seemingly better results so something is a little different.

I also tried my Arduino frequency counter, and although it works fine, just taking the input through a capacitor from the circuit output produces a lot of harmonics, so it needs a separate buffer amplifier. The oscillator prototype itself also needs a buffer amplifier, so I'm going to add that and see how the counter behaves then.

Here is a measurement using the potmeter to adjust the frequency to a precise 1KHz.

Apart from a little bit more noise (six long leads to the potmeter) there is no difference to the THD, so I'll continue with the potmeter installed from now on, and revisit the possible effect on the THD by the tracking tolerance later.

Output amplifier circuit

Because of the loading on the circuit by as little as a scope probe (added noise) and the sensitivity to loading the circuit with an output impedance (adding harmonics), I quickly put together an output amplifier circuit. In hindsight, I should have added that to the prototype, but I didn't...

I am now using the Arduino based counter to show the frequency so I don't need to use my DMM or DSO anymore, reducing the amount of loading and adding noise. With the potmeter at the output, I can now set the voltage level for the E-MU 0202 so it does not show harmonics. The input voltage to the E-MU needs to be around 1Vrms to have the least amount of distortion.

Distortion and the supply rails

Form what I can gather, there is no need to have tracking supply rails with a high accuracy. I could reduce the positive supply by 1V without  a change in the THD. However, the circuit is more critical to the negative supply. If I lower that by 1V, the is a 0.002% change in the THD+N. Lowering both rails to +/- 15V has an even larger effect. I initially thought that the increase of the rails from 15 to 17V was due to the Option2 output amplifier, but I now think that they raised the rails to get a little better THD result.

Power Supply

I've finished a simple textbook separate mains fed power supply for the prototype that has the +/- 17V rails and also the +5V for the counter, and later the relay/reed switch section.

I'm using a toroid transformer that has two separate 24VAC windings.

To get the +/- 17V rails, I'm using the LM317/377 adjustable voltage regulators, and added a trimmer to adjust them. The 5V rail is tapped from the transformer in a way to reduce the loading or digital influence. That voltage is too high for a normal LM7805, so I'm using another LM317 that can have input voltages up to 60V to get the 5V. 

The grounding for this prototype is simple. Later on I will probably split the two windings and create separate supplies for the +17V and -17V, and use a star ground at the output connector for the analog circuits. The digital circuits (the Arduino Nano and the relays/reed switches) will be kept on the 5V rail and that DGND will be as separate as possible.

Running an FFT of the +17V rail with a DC blocker shows that there is a bit more work to be done:

The 50Hz mains comes through, and there is coupling back to the rails from the (1KHz) frequency of the generator.

Granted, it's all just lying on my desk without any shielding from my other equipment (10MHz master clock, and 2 x 10MHz GPSDO) or the quite noisy environment (switching power supplies, DSO, DMM, WiFi router and WiFi transmitter).

After some thinking about the power supply, I decided to make a detour and give the SG505 a more worthy supply. This quick-and-dirty-put-together supply was a failure, so I ripped it apart. 

I have been aware for years of the so called "Superreg" design from Walt Jung from a few decades ago. He designed it predominantly for audio projects, but hey, if it's good enough for critical hi-end audio applications, it should also be adequate for the SG505. At least that is my current thinking.

I don't want to add that project to this blog post, so I'll start a new dedicated one. Superreg

My current test bench

The new SuperReg Power Supply

Details can be found in a dedicated blog post, but here are the first results, using the new supply:

And here are the first results:

This is much more like it, that problem is now seems solved! Kudo's to Walt Jung for a great design.

BTW, I replaced the two 220uF capacitors on the prototype rail input to 22uF versions because the SuperReg does a better job regulating with them. The supply sense leads are soldered directly to the pins of these two capacitors on the bottom of the PCB.

BTW, I don't trust the reported THD+N number very much but I will try to get to that later. Visually, it looks excellent with no visible second harmonic and only a smidgen of a third.

Because I ran into issues with the +5V supply that I added to the SuperReg board, I spent quite some time to get a handle on that. That process can be followed in the post about the SuperReg, towards the bottom.

Now that I have fixed those issues, I can come back here and continue.

Below is the Output Amplifier section that I added earlier, to separate the measurements from the generator itself. I now also added a circuit that creates some separation from the generator output for the Arduino Nano counter input. The transistor circuit is actually a sinewave to square wave convertor, and I also use it to drive the opto-coupler.

And here is the result:

There is a little bit more evidence of harmonics and there is some more mains related 50Hz coming through, but that is no wonder with my current hodge-podge of circuits lying on my desk.

However, there is something strange about the reporting of the ARTA software, the THD looks to be wrong (not even showing here), but the THD+N is in the ball-park. 

On the left is the new SuperReg supply, next to it is the Arduino Nano and the TFT display functioning as a frequency counter, and to the right of that is the circuit with the opto-coupler and the driver circuit. Both boards are lying on top of the E-MU 0202 sound card. The actual output amp proto board connects directly into the E-MU input.

It's actually surprising that this setup functions as well as it does.

If it turns out that the Counter circuit still is responsible for adding harmonics, I will add a switch on the front panel to disconnect it when I need to make really "pure" measurements. Time will tell.

Switching to the REW software

Because I have been struggling to get meaningful data from the ARTA software, I now switched to the Room Acoustic Software Package called


REW version 5.31.3 (REW) I had been using that package to measure room acoustics with a measurement microphone some time ago, and I also found that it supports FFT's.

The calibration information for the software is mostly for setting up room acoustics with loudspeakers and a microphone but I tried the calibration following the online information the best I could. After some testing and trying, I'm now getting this response:

The THD numbers for the individual harmonics and for the total harmonics (0.00019%) almost seem too good to be true. The original specification for the SG505 in the range of 20Hz-20KHz is <= 0.0008% (-102 dB). Unfortunately, they do not specify the THD+N which I measure at 0.0082%. 

I like this software much more and I also think that this is pretty good performance for my setup, so I can now start to check and try a few things and see what difference they make.

I did another run with the Arduino based counter attached, and this is the result:

More harmonics so this calls for a disconnect during further testing, and I will address this challenge later.

Measuring SG505 distortion with a QA304 Audio Analyzer

The Tektronix manual only shows the THD numbers, and there is no picture of an FFT, so it's hard to compare. Fortunately, there is somebody that posts video's on YouTube about this and other units that are of interest. One of them is this one where he measures the distortion of the SG505 with an QA403 Audio Analyzer: YouTube

Here is the result of that tool measuring a SG505:

In that same video he also measures the SG505 Option 2, and the SG502.

And here is the THD report in detail:

Hard to read, I know, but it shows: THD: 0.00027% and THD+N 0.00040%.

This is taken from the output of the SG505 amplifier, and it shows a lot more harmonics.

I'm taking my measurements from the generator isolated with a simple 1x buffer amp just lying around on my desk. THD: 0.00019% (hardly any visual harmonics) and THD+N 0.0084% (a lot comes from the mains related lower frequency range). 

The THD of the prototype is great and I already knew I need to work some more on the noise.

Distortion and the JFET selection

I promised earlier that I was going to try a few different J-FET's and see what the effects are on the generator, more specifically, the harmonic distortion. So far, I had been using the original Tek 151-1021-00 or FN815. I have a few more THT devices and a few SMD devices. I'll start with the TFT versions because I have a socket on the PCB. 

Note that the numbers in the REW rapport fluctuate a bit so I'm using the average number I see in the results below, the numbers in the picture can be a bit different, depending on when I take the screen shot.

151-1021-00/FN815

The original one.

2nd harmonic: 0.000054%, 3rd harmonic 0.000060%
THD 0.00019% and THD+N 0.0084%.

2N4391

According to the Curve Tracer, this J-FET comes very close to the FN815. I have two in a TO-92 metal can version. Unfortunately, I don't know anymore where I got them from. The markings are all on the top of the can and show from top to bottom; PN, 2N4391, m8614. This one may be difficult to get, could be expensive and several variations are no longer manufactured.

First one:
2nd harmonic: 0.00018%, 3rd harmonic 0.000078%
THD 0.00027%, THD+N 0.0084%

Second one:
2nd harmonic: 0.00021%, 3rd harmonic 0.000070%
THD 0.00027%, THD+N 0.0084%

Excellent results.

MMBF4391LT1G

I also have five of these in an SMD package (minimum ordering qty), recently purchased from LCSC.com. I'm just trying one of them.

2nd harmonic: 0.00019%, 3rd harmonic 0.000057%
THD 0.00027%, THD+N 0.0087%

Excellent results.

BF256B

I have a number of them in the TO-92 plastic case. This is not a candidate at all, but just for kicks...

Did not work at all, sinewave does not start.

2N5457

I have two of them in the metal can and had pretty high hopes it would be a candidate.

Did not work at all, sinewave does not start.

J112

Plastic TO-92 package.

First one:
2nd harmonic: 0.00019%, 3rd harmonic 0.000053%
THD 0.00027%, THD+N 0.0083%

Second one:
2nd harmonic: 0.00023%, 3rd harmonic 0.000050%
THD 0.00027%, THD+N 0.0065%

Excellent results.

J113

Plastic TO-92 package.

First one:
2nd harmonic: 0.00021%, 3rd harmonic 0.000051%
THD 0.00028%, THD+N 0.0084%

Second one:
2nd harmonic: 0.00021%, 3rd harmonic 0.000055%
THD 0.00029%, THD+N 0.0084%

Third one:
2nd harmonic: 0.00023%, 3rd harmonic 0.000057%
THD 0.00029%, THD+N 0.0084%

Fourth one:
2nd harmonic: 0.00023%, 3rd harmonic 0.000063%
THD 0.00029%, THD+N 0.0085%

Fifth one:
2nd harmonic: 0.00022%, 3rd harmonic 0.000060%
THD 0.00028%, THD+N 0.0085%

All of the J113's need about 1-2 seconds after power-up before the sinewave starts-up. All the other J-FET's are instantaneous starters.

Very good results.

MMBFJ201

I also have 5 each of these J-FET's in an SMD package (minimum ordering qty) that I'm going to try later on.

Did not work at all, sinewave does not start.

Conclusion so far:

There are several candidates, even some inexpensive ones like the J112 and the J113 with good results, but the best one is the 151-1021-00, of course, and luckily, the 2N4391 and the SMD version, the MBF4391LT1G, also inexpensive, which is even a tad better comes very, very close with excellent results. 

The MBF4391LT1G would be my favorite one.

Distortion and the potmeter tracking

This effect turned out to be more difficult to measure than I anticipated due to the significantly changing THD numbers that are reported by the ARTA software. Changing the value of one of the potmeter halves is more difficult that I anticipated, due to the minute rotation change and the large change in frequency.  To get a better visual handle on this, I used my DSO in the X-Y mode and use the two input channels to look at the output of the two 90 degree phase shift Opamps. It should produce a circle, which it does, but you can't really see enough of a change relative to the changing distortion numbers. In my opinion, it is safe to say that the closer the specification for the tracking is, the better the distortion numbers will be. The overall tolerance of the 10K potmeter value is less important. If needed you could add small resistor values at the bottom and/or top of the potmeters to create equal minimum and maximum values without disturbing the tracking.

I have now switched to the original MOD 100 potmeter took it out of my SG502. I first verified the specifications of +/- 0.5% linearity. Apart from the first several 100 Ohms, this is indeed pretty close:

Setting the resistance precisely is very finicky, you really need the reduction contraption. I have it, but it needs a stable mechanical setup. I had that in the SG502, and either need to butcher it, or create a new setup.

In any case, with it, the results are now: 

2nd harmonic: 0.00030%, 3rd harmonic 0.000054%
THD 0.00035%, THD+N 0.0078%

So no significant changes at 1KHz.

Effect of the timing capacitors on the distortion

I selected the best matching capacitors out of a set of 10 or even 20. For the 1KHz frequency that I have been using so far, I used the 100nF capacitor for the range of 100Hz to 1KHz, using the end of the range. I measured them again, and they are currently both 110.28nF.

As long as they are the same, the actual values do not change the distortion numbers. The other capacitors are 1.034uF and 1.034uF, 9.975nF and 9.975nF and lastly 100.25pF and 100.25pF, so match to well below 1%.

It seems there is nothing to improve here.

FFT with the range selection at 100Hz to 1KHz (end of the range) and the MOD 100 potmeter as a new reference:

2nd harmonic: 0.000043%, 3rd harmonic 0.000073%
THD 0.00016%, THD+N 0.0070%

Switching to the 1KHz to 10KHz range, now using the beginning of the range, the FFT shows this.

2nd harmonic: 0.000049%, 3rd harmonic 0.000096%
THD 0.00018%, THD+N 0.0077%

A little higher numbers.

Here is the 10KHz FFT:

2nd and 3rd are not specified.
Overall THD 0.00029%, THD+N 0.014%

The upward slope above 10KHz is most likely the E-MU 0202 at the end of it's intended audio range.

Output Amplifier

After trying to get the noise and the harmonics further down, I realize now that I need to have a better output amplifier than the simple output buffer I'm using now. There is too much noise injected into the E-MU that I need to quiet down. I already tried putting the whole circuit in a tin can, to no avail.

So, back to the drawing board and figure out what I'm going to use. There are two options, use the standard SG505 output amplifier, or use the Option 2 output amplifier but in a single ended version.

I decided on the more simple solution, and re-build the SG505 output amp.

Here is the schematic for it:

It follows the original schematic, with one difference. The calibration potmeter is now 5K, instead of the original 2K. I tried and tried, but I could not get the calibration done with the 2K. Something is probably reducing the input by another circuit, like the driver for the sync amplifier. 

For the initial tests, the circuit that drives the counter is not connected.

I used the previously used output amp board, and added the new components. To the right of the board is the output level potmeter with a switch to activate the calibration level of either +10dB or 0dB at the output, when terminated into 600 Ohm. 

I'm not sure yet what the exact function is of the JFET in the feedback loop, I've asked Bud if he could shed some more light on it's function. I don't have the original 151-1025-00 part number (listed as an SPF3036), so I tried two different JFET types (2N4391 and the 2N5457), they both seem to work fine and I didn't see a difference between the two.

Eventually, I got it all working according to the documentation, but the harmonic distortion, and the mains related hum is significantly higher plus the noise floor goes up with higher frequencies.  

In retrospect, for this measurement, I did not put the generator in the cookie box, and did not ground the box to the circuit ground, which would have eliminated all the hum and noise. See measurements below.

Bummer, in all aspects this is a failure and a setback. I had high hopes that I could improve matters, those hopes flew out of the window shattering it on the way out. I need to re-think this. Bruce and team would have never designed an output amplifier that makes things worse so something is obviously wrong with my setup.

Revisiting the shielding, using a metal cookie box that earlier did not seem to do much, had a surprisingly positive effect this time:

Spoiler alert, the BNC connector was pressed to the metal of the box by accident. I now have a secure grounding of the box to the circuit ground.

OK, so the left wing is solved by shielding, now what to do about the increase in harmonics on the right wing? 

If (big if) I can convince myself that a proper PCB layout would really solve it, I would go ahead and make one, but I'm not fully convinced at this moment. I need to investigate this a bit more. 

In the original layout of the SG505, the output adjustment potmeter and the calibration switch (blue below) is positioned very close to the output amplifier section, while I have pretty long leads. The potmeter is PCB mounted, and the switch has short wires going to the PCB. BTW, I'm pretty sure that the original PCB has only two layers (I'm missing the usual layer indicators) and definitely has no ground planes on either side.

That potmeter/switch is part of the gain setting, and located in the Opamp feed-back loop. I need to see if that has an effect, but that will have to wait a few days. We have guests and I also have other commitments.

After trying a number of things to get this output amplifier adhering to the specifications, I asked my friend Bud, and he even used LTspice to get some more insights of what this circuit does. Eventually, I asked about this circuit on the diyAudio.com forum and quickly learned about the patent that Bruce got for it. 

Have a look here for an explanation of the patent and that circuit. So that question is answered, but my circuit still does not work right. It may very well be that I'm using an incorrect JFET, and that the circuit is picky about a particular type. The original 151-1025-00 is used in many Tek designs, so my assumption is that it's not an exotic one. The listed equivalent part is the SPF3036, and according to some post about this part, it can be replaced by the equivalent 2N4416. According to the major part houses, the recommended replacement is the J111. I have tried the J112 and the J113 without much success, so I ordered a couple of the J111 that will arrive in a few days.

Using a Notch filter to look at the FFT details

Now that I've arrived at ppm level THD numbers, I need to use a better way to get a handle on the harmonic distortion, and get a better look at changes or improvements. The classical way to help the digitizing analyzer (my EMU-0202), is to reduce the principle frequency by at least 40dB to give the input circuitry of the EMU a better chance of looking at the harmonic details (extend the dynamic range). I have built an active 1KHz Twin-T notch filter many years ago, and more recently upgraded it, so I need to wipe the dust off my audio tester setup.

The instrument at the bottom of the stack is a combination of the Victor Mickevics 1KHz reference oscillator on the left, and the Dick Moore active Twin-T notch filter on the right. Above it is the Pete Miller sound card interface that I do not need for this project, and the top box is of course the heavily modified EMU 0202 "sound card".

With the much higher details that are now required, I needed to put the Twin-T and the reference oscillator in their own metal enclosures, and improve on the power supply. More details can be found on a dedicated blog post here. In short, I now use a string of 9V batteries as the main power source and use a regulator to provide the input voltage. This is followed by a dual shunt regulator already on Victor's oscillator, and I copied his design to the Twin-T. In both cases they supply the very quiet +/-15V rails. The metal enclosures keep the noise out.

It all took a while to sort it out, here is a picture of the mess on my desk while building and testing the power supplies...

Here is the result of the new setup stack:

Twin-T on top, below it the 1KHz Victor oscillator, resting on the E-MU 0202 digitizer. There are no nice looking face plates yet, I don't really need them, that's maybe a project for another time.

So how well does it work now?

Here is the new setup with the Victor oscillator feeding the Twin-T, which has tuned out the fundamental frequency by about -85dB giving the E-MU 0202 DAC an improved range. It shows that Victor's oscillator has a -85.66 dBV harmonic distortion, of mainly the 2nd harmonic, the others are not even visible, which is outstanding. Also note the total absence of mains related hum and the very low noise floor at -130dB. 

BTW, the upward bump of the noise floor left and right of the principle frequency seems to be a "function" or side effect of the Twin-T. 

I think I'm ready now to use this more worthy setup on the SG505 prototype...

Not quite what I expected and hoped for. The prototype with the output amp and the SuperReg is inside the cookie box, the box is connected to the circuit ground and that eliminates all hum and noise. However, it also shows that the harmonics are about 20dBV too high. I can now clearly see them being lifted out of the noise so I can focus on reducing the harmonics or see the effects. BTW, this is with the J113 J-FET in the output amplifier. I tried the J111, J112, J113, 2N4381 and the 2N5457, again, unfortunately, they all have the virtually same 2nd (-81.7 to -81.9) and 3rd (-87.1 to -87.3) harmonic distortion. I have not been able to lay my hands (yet) on the 2N4416/PN4416, the LSK170B or the 2SK1065-5 JFET's.

Just to put things in perspective, below is a measurement on a real SG505 with a QuantAsylum QA403 Audio Analyzer.

The harmonics are about -20dBV lower than what I measured below without the Twin-T. What is of interest is that compared to Victor's oscillator, the SG505 harmonics are clearly visible in both measurements.

As you can also clearly see, the 2nd harmonic is significantly lower with the above measurement than the third. Is that the effect of the correct JFET in the output amplifier?

Even without the Twin-T, the first three harmonics are clearly visible.

Stay tuned, there will be a lot more to come now that I have a fully functioning prototype...

A Github repository is available here it will be updated with information during the project.