Small title bar4

  Home -- article index                                                                    

Most recent update 5-23-2011

Re-greening the Heathkit IG-18: IG-18 #1
Revisiting and improving a favorite project from many years ago

The Heathkit IG-18 Sine-Square Audio Generator (and the IG-5218, SG-5218 and other Heath/Zentih/Schlumberger variants) is a classic piece of test equipment that gave many audiophiles and techies their first experience of a decent quality, wide-frequency-range, low-distortion sine-wave oscillator at an affordable price. I bought and built one when Moses was still alive, then watched through the years as people worked various changes and some improvements on the basic design.

The IG-18's strengths:
1) A simple and elegant design with great economy of parts and generally good-for-the-time performance, designed at the mid-morning of Solid-State.

2) The Bridged-T frequency selector/filter circuit has four decade ranges and uses only five tight-tolerance capacitors to do the work of the 8 that would be needed in a Wien Bridge type of filter, or the 12 needed for a Twin-T type. This design combines economy (tight-tolerance capacitors are expensive) with decent selectivity. The filter only needs 5 caps because the two caps for each range have values that are in a 10:1 ratio. This means that the bridging cap for one range becomes the pillar cap for the next, etc.

Because the caps in the filter have a 10:1 ratio, the notch depth (which becomes a peak in the amplifier response, because the notch is in the negative feedback loop) is about 15-16dB, depending on the exact ratio, or a factor of about 6X (for exactly matched resistors and an exact 10:1 ratio of capacitors, the notch attenuation = 2/(2+ratio) = 0.1666... = -15.6dB, and the peak is then 15.6dB. This formula works for any ratio of caps as long as the two resistors have equal values). This means that the positive feedback loop needs to set the amplifier gain to approximately the same 15.6dB amount in order for oscillation to occur.

3) The use of an incandescent lamp for AGC (automatic gain control), as pioneered by Bill Hewlett, gives really amazing performance compared to other costlier and more complicated schemes -- one simple part replaces many other parts and works about as well. The lamp has a positive temperature coefficient -- as more current flows through it, it gets hotter, and its filament resistance increases. In the IG-18, the lamp is the feedback leg of the positive feedback loop, with the ground leg being a 750 ohm potentiometer that coarsely adjusts the amount of feedback; the lamp's resistance variation with output level supplies the fine control. This maintains the operating point of the system at a good place for both stability and (relatively) low THD.

4) The two discrete frequency selector switches, combined with a ganged-potentiometer vernier, provided frequency selectivity that other commercial gear of the time didn't -- most of them were capacitance-tuned, like the tube-type HP 200CD, with a single dial; they could be finely tuned but generally not quite as finely as the Heath design, where the vernier control gives a total span of 1%.

The IG-18's weaknesses:
As chronicled by Reg Williamson, writing in the Audio Amateur magazine in 1971 (a PDF of the article is here -- search for "IG-18 Mod"), the original IG-18 had many problems. Reg was the first I'm aware of to "green" the IG-18, and his comments and mods led Heath to make some design changes after the article appeared, and to make more changes in the SG-5218. The weaknesses Reg identified were/are (some were fixed by Heath):
1) No "zero" or off position of the vernier control, leading to frequency errors of the output compared to the dial markings (on the old originals, this can be fixed by disassembling the vernier pot and knifing, filing, or dremeling a break in each of the pot's resistance elements near the ccw end, so that the wiper loses contact with the element, making an open circuit). Heath fixed this promptly.

2) Several incorrect part values, including particularly one resistor and one capacitor that contributed to poor distortion performance. These were not fixed for a long while.

3) Insufficient power supply filter capacitance.

4) Early units had a poor grounding scheme for the square-wve output which led to distortion and noise problems in the sine-wave output from the square-wave circuitry.

5) No buffering of the meter circuit from the main output, also leading to rectifier spikes and higher distortion. Heath never addressed this issue.

6) Generally insufficient open-loop amplifier gain, leading to poorer THD.

Reg Williamson's greening (reportedly) fixed most of these problems, and he improved the amplifier circuit by making a Darlington pair for the voltage amplifier/output driver stage Q3, which (supposedly) significantly improved the overall performance by increasing the open-loop gain.

The plan:
I'm going to take an IG-18 and revisit the greening a bit at a time, and then do a bit more, to see how far I can go without completely redesigning the unit (I'm also going to completely rebuild another IG-18 using Bob Cordell's outstanding state-variable oscillator design from the '80s, for which I've done a PC board layout -- I'll build a page for IG-18 #2 when it's finished. The Cordell design is capable of THD levels below 0.0005% (5ppm) at 1kHz. Many have done other designs using some of the IG-18's parts; Morrey's rebuild, detailed in the Audio Amateur, was very extensive but there were reportedly some problems with it as well.

Why do this?
Simple pleasure. Experimenting is fun, and making things better is satisfying. I'm also interested in the results that Reg got -- I think they may be a little too good, given the lamp AGC system and the relatively low-Q filter.

The measuring gear used to evaluate performance and modifications:
-- An HP 334A THD Analyzer, with a distortion + noise measurement floor of about 0.01% and an ability to measure distortion and noise products beyond 5 MHz. This is a very wide-band instrument that does not have any low-pass noise filtering built-in; it does have a 400Hz high-pass filter which is effective for reducing or eliminating hum products.
-- A Core 2 Duo 2.67GHz PC running the ARTA Spectrum Analysis software that has an awesome set of features and controls. Input is through the on-board Intel HD Audio chipset, which in my experience has a THD-only floor of around 0.0018% (18 parts per million) or better, and a maximum sample rate of at least 96kHz and maybe even 192kHz, and 24 bits of resolution. The ARTA software can measure the true distortion only, by computing the RMS sum of the multiples of the input signal, and/or can measure the RMS THD+noise -- very, very nice. This setup will give very good evaluations throughout the audio range, and a bit beyond, with 5+ times the distortion resolution of the HP 334A.
-- A nice Tektronix 7603/7A26/7A22/7B53 100+MHz analog oscilloscope.
-- An HP 3458A 8-1/2-digit DMM. They don't get any better than this.

Test levels:
Output levels will be at a full 10VRMS -- if the unit is stable at all frequencies -- into a 10k ohm load, with attenuator and fine amplitude both full cw (or nearly so) to get the 10V level. Past experience for me has been that with its low impedance output stage, this unit does not care very much about a 600 ohm load -- the output level with the attenuator set to full 10V output drops a couple of volts due to the 150 ohm internal resistor in series with the output attenuator, but otherwise, nothing else changes.

THD levels will also be checked at 1VRMS by setting the attenuator to 3V (+10dBV) and adjusting the fine control for 1V output, also with a 10k ohm load -- this tends to show up hum and noise problems. The bias control will be set for symmetrical clipping of the output waveform or for best stability or for lowest THD, with a best-case balance among the competing alternative settings.

The feedback control will be set for the desired output level with lowest distortion while also maintaining stability at every output frequency from 10Hz to 111kHz. I'm not that interested in frequencies below 10Hz, but if it's stable there, it most likely will be stable at 1Hz too. Then I'll try a feedback setting that gives 5VRMS output to gauge the effect of lower voltage and current on the THD.

The steps:
1) If not already done, fix the grounding for the outputs and level controls, and
   a) Add Reg's emitter-follower meter buffer (albeit with a small change).
   b) Fix the vernier pot if it needs it (it didn't -- this is a later model than I thought).
   c) Depending on spikes present in the THD+noise signal, this may be the time to add a switch to disable the square-wave generator section. It's probably best to just separate the power line for the Schmitt trigger from the sine generator and put a switch in that. One could just switch off the input signal, but that would leave one of the Schmitt trigger transistors in full conduction -- a waste of power.

2) If not done by Heath, and it is needed, increase power supply filtering, then
   a) stiffen regulation by adding a Darlington transistor to the regulator pass transistor.
Possible mods here would be to either
   b) change the value of the zener diode and make the second transistor a voltage gain stage, rather than just a Beta increaser; or
   c) use an LM317. We'll see.

3) If not already done,
   a) change the value of R3 to 56k, add a 500uF or larger cap around R11, and increase C5 to 10 uF (Reg's value) or 100uF (Heath's value in the SG-5218).
   b) Then maybe Beta-match Q1 and Q2, substituting as needed.
   c) Change Q4 and Q5 to higher power devices.

4) Change Q3 to an MPSA64 Darlington transistor (the alternative is to do what Reg did, and splice in a 2N5087).

5) Add a 100 ohm pot in series with D1 and D2 to adjust the bias of the output stage and see if there is any improvement in THD due to lessening of crossover spikes in the output stage. This may require higher-power devices for Q4 and Q5.

6) a) Replace R4 with an N-channel junction FET connected as a constant current source (with a 5k ohm pot from source to ground) to adjust the current for the differential pair of input transistors, or
   b) try using FETs for Q1 and Q2.

7) Do the same as 6) for R12, the 10k ohm load for the voltage amp stage.

First, the original IG-18:
Starting point -- the original design as built by Heath, in a unit recently bought on eBay. Serial Number looks like 09308 but it's hard to read.. Like my first IG-18, this is one of the old "brown" ones with brown and tan knobs, panel, and covers. The black metal end caps and handles that hold the covers on have been replaced with black-painted pieces of wood. The wiring is nicely done and looks a bit more factory than a typical kit, but there's plenty of rosin residue, so it was a kit.

All of the sine oscillator parts are original as per Reg's example -- R1 is 4.7k; C1 is 0.68uF and is a Tantalum epoxy-bead unit. The lamp has the markings SYL 90V -- it appears to be a 90V @ 30mA bulb; a currently available equivalent is the type 90MB with miniature bayonet mount. This is the first IG-18 I've seen or heard of that has the lamp type actually on the lamp. Heath hasn't ever, to my knowldege, disclosed this information.

The Schmitt trigger has a 2.7k resistor added to the bottom of the circuit board paralleling R26, 1.6k, the collector load for Q8, the square-wave output transistor. In addition, there's a 5pF disc cap collector to base on Q8, and a transistor connected as a diode around C8, the 250uF coupling cap feeding signal to the base of Q8. The square wave just doesn't look that great to me, with lots of edge issues and a best risetime of 20nsec. I'm going to undo those mods to go back to square 1.

THD, 1kHz, 10V = 0.12%, independent of load. Nearly all 2nd Harmonic. Contrary to my past experience with two other units, there's no sign of hum or EMI problems. THD, 1kHz, 1V = 0.1%. There is some spiking/notching in the distortion waveform, with some odd order products showing up. No hum at the lower level either. The stability is marginal below 1kHz, but OK above. The feedback control setting is very dicey. All this is most likely caused by the way too-small 0.68uF decoupling cap for the feedback.

Step 1) I think the grounding scheme used by Heath is exactly backwards. For minimum hum and noise, the square-wave attenuator/output ground wire needs to be connected where the center-tap of the power transformer connects, and the sine-wave attenuator and output ground wire needs to be connected to where the square-wave attenuator's wire was connected -- on the ground trace at point L.

With the grounds fixed, the notches in distortion are not from the square-wave generator. I dremelled the supply traces to the Schmitt trigger and ran a power lead over to the sine generator section, disabling the square-wave generator. THD @ 1V unchanged.
   a) Added the meter buffer and spiking disappeared completely, THD @ 10V = 0.1% -- not much improvement. Will have to add the meter damping capacitor at some later point, and possibly Reg's meter compensation network.
   b) Vernier pot fine, with a true zero position.
   c) A switch for the square-wave generator will come later, if ever.

Step 2) Checked power supply ripple with stock 300uF/60V caps; the output voltage is 39.8V, a bit low -- there is no ripple in the output from the pass transistor, but there is 40mV p-p of output signal. coming back from the sine-wave oscillator.
   a) Adding a Darlington transistor made no improvement in supply cleanliness. The fix might be a real feedback-controlled regulator, but not now, if ever.

Step 3) a) Changed R3 to 56k, the value Heath used in the SG-5218 upgrade to the IG-18 (Reg used 47k); I'm not honestly sure what it is this base-to-base resistor does other than damp the Q of the filter somewhat -- it is essentially connected in parallel with the Bridged-T filter plus the feedbck resistors. I increased C5 to 100uF, added a 470uF cap around R11. LF stability improved. THD+noise, 1kHz, 1V = 0.016% with HP 334A; I got the same readings from ARTA Spectrum Analyzer, so the HP is reliable at this level, not that I doubted it. Good LF operation to below 10Hz, then signal just disappears at 3Hz and below; HF operation not yet checked.

Incorporated Step 4 in this step and changed Q3 to Darlington MPS-A64. This worked OK at low frequencies, but there was no stable operating point above 10kHz and the output waveform was very distorted at higher frequencies. I put the Heath PNP back in, and still could not get a stable operating point above 70kHz. Removed the 470uF cap bypassing R11, and obtained full bandwidth with good stability. THD+noise, 100kHz, 1V = 0.12%. Lead dress of the wires to and from the frequency selector and range switches could be causing excessive stray capacitance problems. The MPS-A64 is only rated for 30V and it's possible it was working on the ragged edge. I will try a 2N5087 like Reg did.

Actually tried an MPS6729, which I have a lot of, and used a 100k resistor from its emitter to the + supply, where Reg used 220k. Works well. Tried the 470uF bypass on R11 again -- had to remove it -- with it, distortion was 1% at 100kHz. Removing it gave better stability and dropped the distortion 10-fold. This result is a bit counter-intuitive, given the lower open-loop gain without it; but with no cap, there is local AC emitter current feedback due to R11, the 150 ohm emitter resistor. Perhaps the 2N5087 is the best choice. I have a couple. ...

The 5087 was much worse than the MPS6729 as a Darlington device. I changed R3 from 56k to 100k with an immediate improvement in 1kHz THD and stability. I dug around and found an MPS8599 and replaced the Heath Q3 with it -- no Darlington. This works. It does not benefit from a Darlington second device. It's THD, 100kHz, 1V = 0.03%, the best reading at 100kHz I've gotten so far, and the 1kHz, 1V THD = 0.010%, with reasonable settling time and good stability. Unfortunately, the output level for this performance is 5VRMS max -- any higher and the distortion climbs rapidly. At 5V out, the settling time at 1kHz is long and the lamp is very sensitive to vibration -- other frequencies, higher and lower, settle faster.

Step 5) Replaced Q4 & Q5 with higher power devices, ECG 373 and 374 -- standing no-signal current is 13mA, sufficient to completely prevent crossover distortion, so no bias pot is needed. These devices are overkill to a significant degree. I'm going back to lower power, higher Beta transistors.

Step 6) b) Tried matched JFETs as the diff amp devices. ... The result was OK, but not as good as with the Heath 2N3416s; more noise, more distortion -- just not enough gain. The 56k R4 is a pretty long tail for the diff pair, but maybe it's time to try the JFET current source. ... Found a broken switch wafer on the 1's decade switch, which is why the lowest settings didn't work. Fixed that, which required a major disassembly. Ran the 10V and 1V THD+noise tests again at 10V out: 100kHz, 10V = 0.13%; 100kHz, 1V = 0.11%. 1kHz, 10V = 0.021%; 1kHz, 1V = 0.022%. All a bit better than expected at a basic 10V output from the oscillator.

Have decided to put a 47V zener in the power supply -- everything in the system should be OK with a roughly 45-46V supply output. This may give the voltage amp and the output stage some breathing room. Will at the same time use the original devices for Q4 (2N3416) & Q5 (2N2306) to let Q3 have a bit more dynamic range. These transistors are rated at 500mA and 625mW, so there's plenty of margin for them in this circuit. Not sure why Reg felt he had to use larger devices. Supply voltage is 45.3V. Everything is doing just fine.
-- THD+noise, 100kHz, 10V = 0.067%; 1kHz, 10V = 0.02%.
This level of performance is perhaps 2-5X better than Heath's original -- I have no idea how it compares to the SG-5218.
After tweaking the diff. pair bias control for lowest distortion at 100kHz, the THD at 100kHz = 0.058%; 1kHz = 0.025%.
Readjusted feedback for 5VRMS max. output:
-- THD+noise, 100kHz, 5V = 0.034%; 1kHz, 5V = 0.011%. Half the output, half the distortion, hmmm...
As before, settling time is fairly long at 1kHz, about 3-5 seconds.

See update note on grounding, above... I reconnected the square-wave generator, and made a ground run change -- I ran the attenuator/output ground back to the point where the power supply center tap connects to the oscillator PC board -- the square-wave generator has a negligible effect on sine distortion, especially with the square-wave output controls turned down.

The spectrum analyzer software shows that 60Hz is down about 76dB in the output, and multiples are nearly 100dB down, so hum just isn't much of a problem for higher frequency operation. UPDATE -- I checked the output by setting the range switch to X100, the 1's switch to 0 and the 10's switch to 10, which raises the impedance of the frequency selector by 10X. The 60Hz compnent was much worse, and now I remember the issues I had with my first IG-18. This problem is EMI, not poor supply filtering. The cure in the other unit in which I put the Cordell oscillator circuit was to use a nice Avel-Lindberg toroid power transformer.

Perhaps more could be done with lamp choice -- maybe a lower resistance lamp, such as an 1819 or 1829 in conjunction with a resistor, which would be a less mechanically fluttery solution, leading to better stability and perhaps lower THD. Or perhaps a higher voltage lamp like the SC3 or 120PS (wire leads) or 120MB (miniature bayonet base), or a 4W 120V night-light bulb might work better. I think the thermal mass of the lamp needs to be higher than that of the 90V lamp.

Note that this unit is quite sensitive to the capacitor ratios in the range switch -- a slight change of ratio in range switching easily leads to stability problems and higher than expected level shifts. This is also true for the two resistor values in each pair -- they really need to be matched to better than 1%, but this is less critical than an accurate capacitor value ratio. This is because the resistance ratio changes the frequency slightly, but the capacitor ratio changes the overall gain level and feedback margin. The caps need to be matched to better than 1%. My biggest cap, the 5uF, was 5.05uF, so I used that "505" value for all the others except the smallest, where wiring C comes into play.

Access to an impedance bridge would be helpful here -- the second IG-18 I got just doesn't have a happy place on all the ranges, with big output level shifts and squegging on one range or another, depending on the feedback and bias control settings. Turns out two of the caps are pretty far off.

I think I'm done with this one for now. Good experience, and fun too. I will proceed with putting the Cordell oscillator in the second unit along with a nice, quiet power transformer (see IG-18 #2 on this site). In addition to trying the constant-current source replacements for the resistors I mentioned, another possible mod for the IG-18 is to do what HP did in the 239A oscillator and the 339A test set, both of which use a Bridged-T design -- use 8 capacitors and set the caps in a 100:1 ratio. This gives a notch/peak of around 34dB, meaning that the positive loop needs a "gain" of 34dB too. This significantly enhances selectivity, which should help lower distortion and improve stability -- but the open-loop gain of the amp may not be high enough to allow this with low distortion. Then you're looking at using an op-amp for the oscillator and needing a buffer amp to isolate the oscillator from external conditions, and now you're in a whole different arena, very far from easily improving this old warhorse...

I don't know how Reg Williamson got stable operation at high frequencies using a 2N5087 as a Darlington with Q3, and bypassing R11. Bypassing R11 does improve 1kHz results but significantly increases distortion at 100kHz, and really negatively affects HF stability. Adding the 2N5087 Darlington just made a world of hurt at 10kHz and above. UPDATE -- a post on diyAudio forums from Bill T. indicated that this mod really didn't work and Reg retracted the idea of using it.

As to Reg's 1kHz distortion of 0.003%, maybe Reg misplaced a decimal point? I really can't explain his results -- 0.03% is altogether reasonable, but 0.003% isn't, based on my experience.

If I come back to this unit, I'll try the JFET current sources in place of R4 and R12.

So, the important things:
1) Change R3 to at least 56k -- 100k worked better for me.
2) Change C5 to 100uF/35V.
3) Buffer the meter circuit with an emitter follower! This is so important for low distortion and good high-frequency performance that it can't be overstated. I cut a trace and used the meter cal pot as the emitter resistor, and added a 100uF cap across the cut trace between the emitter and the meter's diode rectifiers. This saves a resistor if you're willing to select the base bias resistor.
4) Check the range caps and pad as needed to get the output frequency to be in exact 10X multiples as you switch the range switch, without changing the frequency switch settings. The smaller, HF caps will be slightly smaller than the others due to wiring capacitance. I used a 470pF cap paralleled with an 8-50pF variable for the X1k range.
5) For lowest THD, set the max. output to 5VRMS with the feedback control. At lower settings of the attenuator and with a 600 ohm load, the maximum output will be 2.5VRMS.
6) Try a current source for R12. Best overall performance will likely be with a current between 2 and 5mA.

UPDATE -- 10-2-2010 I did try the JFET current sources. As I thought, replacing R4 with the FET made no difference in THD at all -- the 56k resistor is sufficiently large to get the job done. Replacing R12 with the FET, however, made a difference. I used a 2SK246, a 50V FET, and optimum performance overall came at a constant current through Q3 (and the output bias diodes) of approximately 5mA, produced by a 120 ohm resistor from source to ground, with the gate grounded and the drain connected to the base of Q5. The bias setting affects THD as well. There is a balance between best 1kHz and best 100kHz THD. Now, with 10VRMS out, 1kHz = 0.016%, 100kHz = 0.068%. Readjusting feedback for 5V output, and resetting the bias control yields 1kHz = 0.014%, 100kHz = 0.028%. Most likely, a different current level through the FET would give best 1kHz THD at the lower output level. Clearly, the current source helps most at high frequencies.
meter buffer

UPDATE -- 10-9-2010
Somehow, in fooling around, I killed the constant-current FET used in place of R12. While I was measuring another FET, which I set a 3.5mA, I decided to replace the diff amp pair with Beta-matched PN2484s as well. This got me to thinking about the amplifier in general -- how best to improve the linearity of the amp without making huge changes, like adding a current-mirror pair to the collectors of Q1 and Q2?

Reg tried increasing the open loop gain by bypassing R11. Through the years, working in low-noise audio gear, it has been my practice to get as much gain as possible in the input stage. I decided to try that here as well. I'm not much good with Spice -- I tried modelling this amp in LTSpice, and I can't figure out how to get a THD measurment with it, or even to show a sine wave on the output. So I plowed ahead with the usual tools -- soldering iron and miscellaneous parts.

My reasonining is that trying to get more gain out of Q3 will inevitably lead to higher distortion, not lower, because of non-linearity. I decided to increase the emitter resistor, R1, from 150 ohms to 1k ohm, which would add a lot of current feedback/emitter degeneration, and also raise the input Z of Q3. Then I increased Q2's collector load from 10k to 22k to get a bit more gain from the diff pair. And, thinking it couldn't hurt, I increased the power supply filters, C1 and C2, to 2200uF each.

The results were actually very good, better than I expected -- 10V, 1kHz THD = 0.016% and 10V, 100kHz THD = 0.056% (HP 334A). And with feedback set for 5V output, 1.1kHz THD = 0.0064% (ARTA). Here's the spectrum at 5V output (I raised the frequency so the spectrum lines wouldn't lay on top of the grid lines):

5V spectrum

These results raised the question of whether further improvement could be gained from raising the capacitor ratio, similar to the HP 239A & 339A. Using 1uF and 22nF would make a ratio of 45 (recall that in the 239A and 339Aoscillators, HP uses a ratio of 100) and put the output near 1kHz, so I used those values on the X10 range. The amplifier had enough gain and the feedback control had the range to get stable operation, but the THD was slightly worse than with the stock Heath 10:1 ratio.

Oscillator parts changed/used: Q1 and 2: PN2484
Q3: MPS8599
D7: 47V 1W zener diode
C1 and 2: 2200uF 50V
C5: 100uF 35V
R2: 22k 1/4W
R3: 100k 1/4W
R11: 1k 1/4W
R12: replaced by N-channel JFET and 220 ohm 1/4W
Make sure that the tuning caps have values accurate to better than 1%. Measure all five, find the one with the highest value, then pad the others with small parallel caps to get the same digits.

Meter buffer parts added:
Qm: 2N3904 NPN
Cin: 10uF 25V
Cout: 100uF 25V
Cmeter: 100uF 25V
Rb1 and Rb2: 1M 1/4W
R108: 47k 1/4W

Adjustments Setting the bias and feedback controls worked best for me this way: Set the oscialltor to 100kHz. Fiddle with the bias and feedback pots until you get a stable 100kHz output of 10V with the attenuator and output level controls full up. Once 100kHz is good, then check at 1kHz. If you have a distortion analyzer, you can get a small improvement by slightly adjusting the bias control.

For lowest distortion with good stability, set the feedback control for slightly more than 5V output, then check the 100kHz output for stability and tweak the bias control if needed. You might be able to further improve THD by setting the output for 3V, but stability and settling time may be problematic.

The new combination of parts values has led to better overall stability and settling time is now acceptably short at all frequencies at 5V output.

Given the relatively low sensitivity of the lamp feedback, I think this performance is better than good. So three transistors and a large handful of parts has got the IG-18 working very well indeed. Not state-of-the-art, but real good.

12-9-2010 update
I hadn't turned this oscillator on since I finished the upgrades. I found a couple of things -- it didn't start up at 1kHz, and its 100kHz output was erratic. Moving the range switch fixed the startup -- as soon as I moved the knob, the oscillator started. Then I had to adjust the bias control to get a good 100kHz signal. The start-up problem persists over all settings of the various controls -- I have to move the range switch to get the oscillator to start up. Once going, it's fine. I can live with that, although I'd like to know the cause -- usually the transient when power is applied will start the oscillator. It may be that I've filtered the power supply too well and the start-up transient has too low a rate of change.

The other day I received an HP 8903E THD analyzer. This nice old unit has fully automatic operation, with auto level control and auto tuning, and a resolution of around 0.0015% 0.0009% at 1kHz, all of which is very convenient when you're fiddling with the feedback and bias controls. With it, I discovered that the setting of the feedback control didn't have an enormous impact on THD, but did, naturally, have a huge impact on output -- with feedback set for a maximum output ranging from 5V to 10V, feedback went from a low of about 0.008% at 5V to a high of about 0.012% at 10V -- so call the 1kHz THD 0.01% at max ouput.

What I did find is that the THD is very sensitive to the settings of both the 10dB step level control and of the variable output vernier control -- seems clear that a buffer amp on the output would be helpful. I also found that 60Hz hum in the output is a real problem at the lowest settings of the tens frequency switch, where the bridge impedance is highest. No surprises about the hum, but it is a bit discouraging if you want a good quiet oscillator. Fixing that will mean either replacing the power transformer or -- maybe better -- moving it out of the oscillator to an enclosure a few feet away.

5-20-2011 update
Recently I received an email from Richard Andresen, the author and publisher of the fine circuit simulation program 5Spice. Richard was curious about some modifications he had in mind, so we began exploring a few further mods. The idea was to get more open loop gain without sacrificing stability, with the end goal of lower distortion, especially at higher frequencies. We decided that Richard would work on a recently acquired IG-18, check for large problems, and if all seemed well, I would duplicate the mods as a check on variability, and then I would use my high-resolution distortion measuring gear to make a final evaluation of performance.

This work is now in progress. I hope to have more to say soon.

5-23-2011 update
BTW, here's an update to ther attenuator switch that uses the 1% resistor values from the HP 239A -- a very good fit for the IG-18 in any incarnation, and using the internal 600 ohm load switch to switch the ground between a floating ground coupled to the chassis with the Heath 47nF cap and a hard chassis ground is very useful. Note that the load on the output of the oscillator is a minimum of 1200 ohms if the 10V output is terminated in 600 ohms -- this load is OK for the IG-18 although it does increase THD somewhat. But into a high-Z load, no problem -- and the output Z is 600 ohms at any setting of the attenuator switch.
IG-18-1 attenuator

back to top

to Home -- article index

© 2005-2011 Dick Moore