IG-18 #3, the IG-339A

Re-written 5-21-2012, updated 6-11-2012

Re-greening the Heathkit IG-18: IG-18 #3
The IG-339A oscillator, an HP 339A oscillator in new old clothing

If you've visited this page before, you know I intended to describe the modification of a Heath IG-18/SG-5218 to significantly improve distortion performance. That project is now ~~well underway~~ completed. This page is almost as much about measurement and analysis as it is about modding an old Heathkit to work better.

UPDATE 6-11-2012
Where is the IG-339B mentioned on this page, below?
I've created a separate page for the B version of this oscillator. The IG-339B uses the 10:1 capacitor ratio of the Heath IG-18's range switch, which somewhat simplifies the construction of the oscillator. Click the link above to the B version page.

Simple magnetic shielding for the power transformer
Please note the end section of the B version page that deals with magnetic shielding of the power transformer. That shielding will also help this A version build a LOT.

Fixing a previously unnoticed and peculiar noise problem
I discovered a signal contamination problem by accident, and it was serious -- my PC somehow interferes with the Active Twin-T notch filter and the E-MU 0204 USB sound module by generating noise spikes starting at exactly 1kHz and proceeding up in frequency at 1kHz intervals. These noise spikes were hiding the true levels of 1kHz test signals and especially the harmonics at 2kHz, 3kHz, etc. The solution was to physically separate and re-orient the various boxes -- this makes my measurements much less convenient, but solves the noise spike issues.

What can you expect from this build?
I've edited some spectrum plots for easier viewing to compare the performance of the HP 339A, HP 239A, and the current IG-339A. The IG-339A's performance sits comfortably between the 339 and the 239 at 1100Hz. I moved the test frequency up slightly from 1kHz in order to be sure that I didn't have the signal and noise contamination discussed above.

Several things to note here -- the line frequency harmonics in the 339A and the IG-339A are spaced at 120Hz, while those of the 239A are spaced at 60Hz. I don't know exactly why; the 239 is the quietest of the three at these frequencies. All three spectra were taken under the same conditions of output level (6.6VRMS), sampling frequency, and FFT sample size. I used the Active Twin-T notch filter to lower the fundamental level to -60dBV so that the calculated THDs can be compared. A little of the 1kHz-spaced products can still be seen in all three plots -- greatly reduced, but not eliminated.

I had modded the HP 339 a while ago with a pot to control the cancellation of the 2nd H. in the AGC JFET, the same as has been done to the IG-339A. I have not made this mod to the HP 239 yet. The 2nd H. cancellation is not as good in the IG-339A as in the HP 339. The two 339 variants have much lower 3rd H. than the 239 -- this may be due to the small design differences between the AGC systems of the 239 and the 339.
###

Introduction
Quite a while ago, my online friend Larry Burk sent me a link to Steve Lafferty's IG-18SL mod project page detailing his mods to use the circuitry of the oscillator section of the HP 339A Distortion Measurement Set in the IG-18. I'm going to call my unit the IG-339, with two versions: IG-339A and IG-339B.

I had previously written in 2011 that I was planning on modding an IG-18 to the circuitry of the HP 239A -- the 239A and the HP 339A oscillators differ only in some details of the automatic gain control (AGC) loop. But they are essentially identical and also measure nearly identically. So when I looked at Steve's great article, I was very interested in how far we had been going down the same road, including re-wiring the power transformer primary to get the right secondary supply voltages.

Then Larry told me he was etching and drilling a few circuit boards for Steve's project and offered to send me one, That's when I decided to use the 339 circuit instead of the 239. I really hate making circuit boards. Many thanks to both Steve and Larry for their good work. Please read Steve's pages carefully, since they have very good and clear descriptions of the changes to the IG-18.

On my other IG-18 webpages, IG-18 #1 and IG-18 #2, I modified Heath IG-18s bought cheaply on eBay. I made a few relatively simple mods for lower distortion in IG-18 #1, and made much more extensive mods for very low distortion in IG-18 #2.

Both projects ultimately were successful in achieving their goals. But evaluating those results requires some very accurate analysis equipment. The HP 339A, for example, cannot resolve the distortion of its own oscillator -- it has a measurement floor of about 0.001% which is the result of a relatively high level of 2nd Harmonic distortion in the notch filter circuitry. So how can distortion be determined?

Test equipment for this project
A high resolution spectrum analyzer is essential, whether a software analyzer or a hardware analyzer. I use the spectrum analyzer function in the ARTA Audio Analysis software, which runs under Windows on a PC. Input to the PC and ARTA now is from an E-MU 0204 USB sound module. This unit has 24-bit ADC and DAC converters and a maximum sample rate of 192kHz, so it serves for my testing at up to 15kHz fundamentals with analysis of harmonic distortion products up to the fifth. It also has extremely low self-distortion, potentially under 0.0002%. The EMU is fed from a level control pot driving an Active Twin-T notch filter, which removes the fundamental input signal without significantly altering the harmonics in that signal. This improves the dynamic range of the distortion measurement to more than 130dB for audio band signals.

Any measurement errors?
Nothing important, once I cleared up the mystery about spurious signals and noise issues connected to the relative placement of the various parts and pieces of the test equipment and the computer. Please see the linearity information on the Active Twin-T notch filter page.

As I've noted on other pages here on the site, higher harmonics than the 3rd are generally very small in very low distortion oscillators, so the roughly 90kHz upper frequency limit (approx. 1/2 the sampling frequency) of the EMU is not a concern. However, noise is, and the noise spectrum of the EMU begins to rise significantly above 40kHz, so resolving higher harmonics above that can be problematic. Finally, I use the Active Twin-T notch filter for highest resolution by notching out the fundamental while leaving the harmonics essentially untouched. Additionally, the extra 20dB of gain available in the filter helps to raise the distortion products above the noise. Given the very low self-distortion of the EMU (after you carefully optimize the input level to take full advantage of the resolution) you may wonder why you need the Active Twin-T. You'll see shortly.

So, why IG-18 #3? If we can get close to HP's level of performance with a modded IG-18, well, that's just great news. And what level of performance would that be? Here's a spectrogram of the output of the HP 339A oscillator at 1kHz at full output of 6.7VRMS (mine runs a little high -- cal should be 6.32VRMS for the essentially unloaded output), into the medium input Z of a level pot and the Active Twin-T notch filter. The 0dBV reference level is 1VRMS, and the input signal has been set to 1VRMS with the level pot used with the Twin-T filter.

I tuned the notch filter for 60dB of attenuation of the fundamental, so that the spectrum analyzer's calculated values for THD would have a meaningful reference level. To get the actual percentage levels, divide the displayed percentage by 60dB = 1,000. This means that the actual RMS THD level is 0.074%/1000 = 0.000074%. That's why you need the Active Twin-T. BTW, I prefer a linear frequency scale for distortion plots because it really helps to find products in a lot of noise if you know where they should be.

Note that the 2nd H. is buried in modulation artifacts and noise, and that the 3rd and 4th are around -130dB. The THD+noise level is higher because of the presence of line-frequency products and because of the actual shape and level of the noise curve at low frequencies. I generally disregard the THD+noise figures because of the spuriae that don't have anything to do with actual non-linearity.

I don't care what you compare it to, this is just an excellent oscillator. And the HP 239A offers essentially identical results. It's a shame that the analyzer section of the 339 doesn't even come close to the performance of the oscillator. Ideas, anyone?

IG-18 on steroids
Oddly enough, you can think of the oscillators in the 239 and 339 as more modern versions of the IG-18. All three use a Bridged-T bandpass filter for frequency selectivity. All use two banks of switched resistor decades and a dual potentiometer vernier for frequency control. All have four ranges and all span a total frequency range from <10Hz to >100kHz. All use relatively simple amplification schemes. But then the differences begin to appear.

The range capacitors in the IG-18's Bridged-T frequency selector/filter circuit have a 10:1 ratio between the bridge and pillar positions -- please see the IG-18 #1 page for full details.

But in the 239/339, the capacitors have a quite large ratio of 100:1, which means using 6 capacitors in the 339 to do that work of 8 (as described in IG-18 #1). And in the 239, there are actually 8 caps, which makes the push-button range switching used in that unit's design simpler. The 100:1 capacitor ratio means a filter notch and resulting amplifier peak of 34dB, and that means much more gain and bandwidth are needed from the amplifier, especially at 110kHz. But the benefit is higher filter selectivity, or Q, which means higher frequency selectivity, resulting in higher rejection of distortion products, which in turn means less distortion.

The 100:1 ratio is about as high as I would want to go, given the resulting very wide spread of capacitor values -- and to also yield reasonable values that work with the existing Heath bridge resistors. In the 239/339, the caps range from 5.6uF down to roughly 56pF (roughly, because of stray C effects). HP chose to use larger resistor values that require smaller cap values in order to minimize the loading on the HA2625 amplifier by the bridge network. With newer opamps like the LT1468, OPA1641, and LME49710, this wouldn't seem to be a problem, so I'm keeping the range resistors as-is in the IG-339A, and using larger caps.

The HP 339 AGC system uses a half-wave peak-detector acting as a sample-and-hold, an amplifier, an integrator, and an N-channel JFET voltage-controlled-resistor (Teledyne VCR2N) as a variable resistor in the AGC feedback loop for level control. The higher gain of this AGC system yields much tighter amplitude control and better operating point control, and therefore lower distortion than the lamp system of the IG-18 can achieve. An old friend used to call this kind of improvement "hanging a barn door with electronics." Gotta love the lamp, until you want ultimate performance.

Some general construction notes
Given the IG-18 chassis, power transformer, and various switches and controls, and a proven working PC board, either version is worth building. These days, it's the mechanical stuff that's hard to find.

In my other web pages about the IG-18, I cast doubt on the Heath power transformer, accusing it of being a source of hum and noise. This is not the case -- it is very quiet, although line noise can be reduced in this circuit; but there is no need for a different power transformer unless you need to run off 240V line power.

I'll use Steve's part IDs just for the sake of simplicity. I chose to physically turn the power transformer around 180° so that it's secondary wires can reach the PC board without splicing in more wire. This may mean moving a bit more of the power cord into the chassis, depending on the length of the existing cord from the Heyco retainer to the other wiring.

Here's a pix of the PC Board with some parts called out. Note the three resistors in the middle, next to the feedback pot, mounted in socket pins -- these are the feedback leg and ground leg resistors associated with the JFET. The 1000uF power supply caps I used just barely fit. The blue "bulbs" are Tantalum caps. Note the dual-to-single 8-pin DIP socket for U4:

For the IG-339A, I added two decks to the range switch and used eight caps for the range switching, and built a separate board to hold the four larger pillar position range caps and one set of the AGC filter caps. See the detail pix below, in the Range switch section.

Power supply -- I increased the four power supply filter caps from 470uF to 1000uF, and changed R1 and R2 from 36 ohm 1W to 100 ohm 1/4W. I intend to use a separate +5V regulator for the square-wave section, which results in lower loading of the +15V supply. The higher value caps and resistors mean less ripple at the inputs to the +15 and -15 regulators, giving them less work to do in ripple rejection. Although I didn't do it yet, I recommend using LM317 and LM337 regulators for the +15 and -15V supplies -- especially for the negative supply. The 7915 negative regulator just does not perform anywhere near as well as the LM337 used with the appropriate resistors to set the output voltage. It's noisier and has worse ripple rejection, not to mention poorer regulation.

The rear panel of the IG-18 makes a very good place to mount the regulators, but be sure to insulate both/all of them from the panel to avoid grounding problems.

Parts HP didn't use -- Steve didn't indicate why he chose to isolate the drive to the AGC rectifier with a unity gain buffer -- the HP 339 and 239 don't use it and obviously don't seem to need it. But it doesn't hurt anything either, since this is not AC feedback but DC. But it is essential that the rectified signal not have a low-Z discharge path, so U4B really should be a FET-input opamp, so that the filter Cs 21-24 only discharge through the 5.1M R36. Steve makes this amp half of a dual and calls for an OPA2134 -- an excellent choice, but note that 1/2 of this amp is the output buffer, and an LT1469 may be better -- but there are issues -- I'll talk about this later.

Range switch -- The Heath range switch is a lot longer than it needs to be, which is good, because the easiest way to add the switching for the AGC filtering is to add two decks to the range switch. A handful of small spacers (brass, aluminum, or plastic tubing -- see your local hardware store) and a couple of decks scavenged from another old switch or two, or taken from new ones, will let you easily mount the needed decks. I had two complete Heath range switches, so I just took the decks from one and added them to the other. Most people won't have that option, but the switches that Heath used are a fairly common size with 30° rotation detents, which is the usual amount for many, many switches.

Here's a pix of the modded range switch as a guide. The four large pillar caps (3 caps for each one) really take up space. There are only 3 caps for each of the AGC filter sets, because one of the caps for each set is mounted on the PC board:

Cap values -- The biggest difficulty for most constructors will be selecting cap values. The needed caps are in the "16" range -- all are very close to 16 in value, with different decimal points. These need to be accurate to 1% or better. They are fairly easily made from parallel combos of 15 and 10 series caps. All of my pillar section range caps ended up being combos of three parts to get the accuracy and matching needed. A reasonably accurate cap meter or impedance bridge is essential -- the range caps must be selected to 1% or better for good results and stability, since any changes in the cap ratios mean changes in the circuit gain. An inexpensive 3-1/2-digit digital cap meter will work fine, since absolute accuracy is not so important, but ratio accuracy is. I use both a digital meter and an old but very accurate ESI 250DE RCL Bridge.

The IG-339A runs
I finished the main board and switches and made a board to hold the bigger film caps for the pillar portion of the bridged-T and for the film cap input filters for the AGC rectifier. The smaller bridge caps for the bridged-T are on the first deck of the 4-deck range switch and the AGC integrator caps in the U3A circuit are on the fourth deck. You can see how well the turned-around power transformer works out:

Then I powered it up and it oscillated at 1kHz. Of my two JFETs that work, a J105 and a 2SK152, the 2SK152 seems to be the best, and I was able to get decent operation at 100kHz, letting me adjust the trimmer cap I used in parallel with a 100pF silver mica cap to make the 160pF smallest tuning cap. But neither of these JFETs are the right parts for this unit. Larry Burk pointed me to Electronics Goldmine, who have some of the Teledyne VCR2N JFETs used by HP, which have a nominal resistance range of 20 ohms to 60 ohms -- should be ideal for this circuit, so I ordered a few. I also ordered some PN4091 and PN4092 JFETs from Mouser to see if one of them will work -- Steve Lafferty used the 4092. That's the good news.

The bad news is that gain settings that give stable operation across all ranges also have distortion a little higher than desired -- around 0.0005% or higher, depending -- but not even in the same ballpark as the real 339. The bipolar opamps LME49710 and NE5534 do not do well for U1 in this circuit -- I suspect their low input Z loads the bridge, but I'm not sure that's the cause. Just for grins, I plugged in an HA-2625 for U1 (used in the HP 339 and by Steve) but without any of the compensation parts, and it was really bad -- it had OK level control and fairly stable operation across all ranges, but tons of distortion -- this is altogether strange -- it wasn't oscillating and clearly was in a linear part of its operating range. The FET-input OPA1641 and the bipolar-input LT1468 work the best, with the OPA134 close behind. I initially used an OPA1642 (dual 1641) for U4, the output buffer and the AGC difference amp. I really don't know now why the FET-input amps work all right, if not great, but the bipolar-input amps have very high distortion. Mystery.

This would be a good place to mention the possibility of using an OPA1641 or an LT1468 as an amp whose low-Z output can provide a virtual common ground for the positive and negative feedback legs of the bridge, in the way that Jim Williams used one in a Wien bridge oscillator in his Linear Technology App Note 43, Fig. 47 -- please note that Fig's 47 and 48 are swapped -- 47 is the right one. This amp potentially greatly reduces common-mode error, and may be a very good idea.

Parts arrive and results are good
Subbing in the VCR2N, PN4091 and PN 4092 revealed that the VCR2N and the PN4091 produce essentially identical results, and that the PN4092 works the best of the three. This is a little surprising but it's an OK result. Adjusting the feedback pot and the 2nd H. suppression pot (this pot is not in Steve's version -- it's been added by Larry and me) gave very good results at 1kHz. Then I added a 200pF cap around R50 in the AGC loop -- this resistor is 2k ohms in my unit, same as the HP value -- and that gave very good performance on all ranges.

Which opamps?
It was time to find out which amps would give the best overall results. All the bipolar amps except the LT1468 had already proven to be poor performers for some unknown reason, but I knew the FET-input amps all worked. Which amp works best? I swapped in various combinations of OPA134, OPA2134, OPA1641, OPA1642, and also the LT1468 -- I don't have any duals of this part (LT1469), but I do have a few dual-to-single converter sockets, so I could use two LT1468s for any spot where Steve used a dual.

The results were clear -- the LT1468 is the best for U1 and for U4A, the output buffer amp. This was especially true at 100kHz, where my HP 339 measured the oscillator at 0.006% THD (-84.4dB) -- I don't have a hardware spectrum analyzer, so I couldn't make a high resolution measurement at 100kHz; but I have reason to think that the distortion is actually lower than that. My final opamp arrangement is:
U1 - LT1468
U2A & B - OPA2134
U3A & B - OPA2134
U4A - LT1468 or LME49710
U4B - OPA134

It probably would be most convenient to use a dual part for U4A & B, and the LT1469 may be fine there unless the lower input Z of this bipolar part causes problems at low frequencies due to bleeding the AGC input filter too quickly -- but it does work fine on the X10 range and above, and it seemed to work OK at 100Hz. The fallback is to use an OPA1642 or OPA2134.

Here's the spectrum at 1kHz, with 20dB of gain in the active Twin-T filter; this means that -100dBV is actually -120dBV. You can always ignore the cursor data in these plots -- I usually don't set it or use it, and it's just there. As you can see, the harmonic distortion products are all under -120dB, all under 1ppm:

And here are the results at 10kHz, all the same as above except for the sample rate, and including the extra 20dB gain. This means that the 2nd H. is at -111dBV, under 0.0003%, or 3ppm:

The 60Hz spike is at around -113dBV and the other line-related products are very small, which also reduces the modulation effects. I can't explain the 2kHz artifacts, all at the same level -- weird.

IG-339A wrap-up
The performance shown above results from a gate voltage on the PN4092 AGC JFET of about -0.7VDC at 1kHz, with an AC output of 6.75VRMS. This gate voltage will vary a lot with range switch settings, but should never be less than about -0.1VDC, and probably not more than -1.2VDC. The output amplitude variation is within +/- 0.05dB, but might vary a bit more, again depending on a variety of factors. I think it should always be within +/- 0.1dB -- but be sure your AC voltmeter is accurate at the frequencies being measured!

Because the output is metered, I didn't feel a need to accurately set the output level to 6.32VRMS (or anything else). If you want a different output level, you can vary either R34 or R35 to change the "reference" level that determines the output. Reducing the output level may result in lower distortion, particularly at 100kHz -- maybe.

I briefly considered turning the output attenuator switch so that the max value is 3V (the 600 ohm terminated output level) and then setting the meter to full scale, and then making the lowest position an "off" position, but I decided that was overkill. But it would be easy to do. Alternately, you could make the unity-gain output buffer a X1.5 gain stage, raise the supply voltages to +/- 18V, and get a full 10VRMS output. This would be more attractive to me. All of the opamps used here can handle the higher supply voltages.

I used lowly 1N914s for the AGC network diodes and not the 1N4148 and Schottky diodes that HP and Steve used. Larry is sending me some Schottkys and I will try them when they arrive and see if they matter very much. I suspect that if they matter it will be at 100kHz, but since I don't have a hardware spectrum analyzer, it will devolve on others to figure out if they are really important.

I set out to try to answer the question of whether the performance of the HP 239 and 339 oscillators, which used the now-scarce HA-2625 opamps, could be matched using modern, readily-available, unity-gain stable opamps -- the answer is yes, and those amps are the LT1468 and the OPA1641. I've further established that the PN4092 works very well for the AGC JFET, and those are still available, too, although they are in limited supply. Once they are gone, some other form of AGC control or a totally different topology will be needed to equal or better their performance. I really wish I had a way to see high-resolution distortion performance out at the harmonics of 100kHz -- a lot could be cleared up.

Output attenuator mod
Here are the details on the output attenuator. This mod is not essential -- unless you want accurate attenuation and a known and accurate output impedance. All resistors are standard 1% values. This attenuator gives an output Z of 600 ohms at all settings, while presenting the unity-gain output buffer with an acceptable load even when the output is terminated in 600 ohms.

The triangle ground point and the top wire of the floating/chassis ground switch are connected to the ground wire on the attenuator switch that returns to the "clean" ground on the PC board at point M. The S1 switch is the re-purposed front-panel impedance selector switch for internal or external 600 ohm load. I used the Heath 47nF cap and left the chassis ground point of the cap where Heath put it, under the Z switch mounting screw, and soldered the bottom wire from the switch to the same solder lug the cap is soldered to. Seems to work fine, but I may run it's wire from the S point on the PC board. At the moment, line noise is low, well under -100dB.