MUFF WIGGLER Forum Index
 FAQ & Terms of UseFAQ & Terms Of Use   Wiggler RadioMW Radio   Muff Wiggler TwitterTwitter   Support the site @ PatreonPatreon 
 SearchSearch   RegisterSign up   Log inLog in 
WIGGLING 'LITE' IN GUEST MODE

2164 analysis, and comparison to OTA
MUFF WIGGLER Forum Index -> Music Tech DIY  
Author 2164 analysis, and comparison to OTA
guest
this work got started due to this thread:
https://www.muffwiggler.com/forum/viewtopic.php?t=191492

and i wanted to start a new topic to contain all the info ive been gathering along the way. i will keep updating this first post to keep all the information in one, easy to read spot.

there are only a handful of multiplier chips that are commonly used. of the ones that are available today, its the V2164, THAT2162, LM13700, AD633, and MC1496. the 2164/2 are exponentially controlled, and the remainder are linear. the 1496 (single) and 13700 (dual) are the cheapest of the bunch, around a dollar in single quantities, and dropping to under 0.5$ in the thousands. the V2164 (quad) can be had for around 4$ in single quantities and under 2$ in the thousands. the AD633 (single) is the most expensive at 11$ singles and 6$ thousands.

i see the 2164 and 13700 used a lot, so i wanted to get a good understanding of what the tradeoffs are between them. as a starting point, i wanted to replicate the THD+N graph from the SSM2164 datasheet:



this is an updated version from the previous thread, fixing a few things that were wrong. the OTA here has a current source driving the linearizing diodes, and is trimmed for minimum bleedthrough. the 2164 isnt trimmed at all. the output for the OTA is shown in two different positions, depending upon how much headroom you want it to have, as it clearly does not have the dynamic range of the 2164. the right curve (OTA) has good distortion performance, but quickly runs into the noise floor, whereas the left curve (OTA-Shift) has moderate distortion performance, with little headroom and adequate noise floor.

the straight lines on the left hand side of all curves are the noise floor coming up. the "no mode" resistor version is just starting to get into the noise floor at -40dBu output. you can decide the tradeoff between noisefloor and distortion with the mode resistor. here is another graph showing just the SNR (larger values are better) as a function of output signal:



the noise in both the OTA and 2164 is mostly determined by the amount of current in the output stage transistors. a lower mode resistor increases this current, increasing the noise. for the OTA, this noise level is fixed, and very large, as the current in the OTA is set as the maximum output drive level. but, in the 2164, this current is set at a very low baseline via the mode resistor, and then increases as a larger signal is needed. this is why the OTA SNR rises linearly, whereas the 2164 flattens off at the top. at some point, the added current to create the output signal is much larger than the mode current, and signal and noise track equally. so there are diminshing returns to running the 2164 greater than 0dBu with no mode, and +10dBu with 10k mode. for the OTA, you always want to run it as high as your distortion requirements will allow.

[ADDITION: 12/29]

there is a large operating space for both of these ampifiers in terms of input signal and gain. the graph below tries to show the SNR for this space, and is sadly the best of all options ive been able to come up with.



there are 4 sets of clustered curves. each cluster contains the 2164 with no mode resistor, with a 10k mode resistor, and an OTA (except the top cluster which is beyond the operating range of the OTA). for all of these, the OTA is analyzed in the "shift" range as explained above, where 150uA of bias current equals 0dB of gain. this is the OTAs sweet spot for SNR. there is no increase in SNR performance for running an OTA above 500uA (as can be seen in the graph above - more than 10dB of gain and the SNR flattens out), and running an OTA above 250uA incurs heavy distortion penalties (due to emitter resistance). so 150uA gives you +6dB of gain before distortion and SNR that is only a few dB down from max possible. the input signals are dBu into a 30k resistor for the 2164.

for all of the amplifiers, the SNR gets better as you increase the gain. this is due to current noise rising with I^0.5 whereas the signal rises with I. for the OTA there are limitations to this which i believe are due to base resistance thermal noise, and in the 2164 there are limitations due to the signal generated noise as described above. this can be seen in this graph of the noise floor of the two topologies versus bias current:



after 150uA the OTA noise rises linearly with current, whereas the 2164 still rises with I^0.5 (again, uncertain as to why, except for possible much smaller base resistance). the 2164 flattens out at the bottom due noise contributions from non-output-transistor related sources (like its internal OTA and the 30k output resistor). so for very low gains, the OTA actually performs better than the 2164 for SNR. this can be seen at the -40dB gain points on the graph above. but, these regimes are not very useful, as the overall SNR is not as good as at higher gains.

its also worth noting that the relative advantage of going with no mode resistor for SNR reduces greatly for high input signals and low gain settings. so, if you are modulating the mode current to optimize the tradeoff between SNR and distortion, it might be worth incorporating some of the gain signal into that mix as well.

one thing i noticed from taking tests, was that the distortion in the 2164 had a lot of harmonics all the way up to, and past, 20kHz. the OTA, on the other hand, really only had 2nd and 3rd harmonics. part of this is the noise floor being higher on the OTA, so you cant see the lower level harmonics. but, i thought id go back and take measurements for the full range, as compared to just 2nd and 3rd as the datasheet did. what i found was that the distortion was worse for the no-mode resistor condition, and not much different for the 10k mode resistor condition.



above is THD (but not noise) across the full audio bandwidth for the no-mode resistor and 10k conditions. this was done across control input gain, from +20dB to -40dB. the main thing to note, is that the horizontal axis is input signal level for gains of 0dB and less, and is output level for gains above 0dB. the reason it is done this way is because the input and output stages of the 2164 are symmetric, and so are their distortions. lining them up this way shows this pretty clearly, as +20dB matches -20dB, etc. the upturn at the lefthand side of the curves is where the distortion disappeared into the noise floor, and i couldnt measure it anymore.

you can see that the 0dB condition is the best for distortion, and that it moves equally away from this for positive or negative control signals. this distortion level maxes out pretty quickly, so its best to think of the non-0dB curves as normal operating points, with 0dB being an optimal null point. this is more true for the 10k mode resistor condition, as the no-mode resistor condition doesnt vary too much, although the +12dB null goes away as you stray from 0dB gain. so, although it might be tempting to design an amplifier around that null condition with no mode resistor (as it has low distortion and low noise), it will quickly regress back to high distortion as you change the gain. im not sure whats up with the -40dB gain no-mode condition. im thinking the data might be in error there. i expect it to be worse, but not that bad.

now, for a comparison to the LM13700. for this graph, i just used the 10k mode resistor condition, as the distortion is so much worse for the no-mode condition that its not a useful comparison.



this graph is THD+N by input level for both the OTA and 2164, with each having the same colored lines for same gain settings. the OTA is the dashed, whereas the 2164 is solid. again, the OTA is shown in the "shifted" regime, and has limited headroom in comparison to the 2164.

because of the input/output symmetry of the 2164, the distortion is much worse as the gain is increased for a given input level (output level rises as gain rises, increasing distortion). the gain does not have as large of an effect on the OTA. its distortion stays relatively constant from +10dB to -40dB (the curve gets pulled up by the noise floor fro the -40dB condition). at +20dB gain, emmitter resistance dominates, and distortion becomes much worse.

again, the 0dB condition is a sweet spot for the 2164, but for deviations from this, the distortion becomes much more comparable to the OTA. in particular, for positive gains, where the OTA performs better up at +20dB. the results of the SNR analysis can be seen to be repeated here, as the lefthand side of the curves favor the OTA for low gain levels (-10dB and below), and favor the 2164 for high gain levels (0dB and above). this low gain noise advantage is minimal for the OTA, but the 2164 is much better above 0dB. so, to sum up, at higher gains, the OTA has better distortion performance but worse noise, and at lower gains the 2164 has better distortoin and worse noise.

[ADDITION: 1/4]

and now for bleedthrough. ill just cut to the chase, heres the graph:



on the left is the bleedthrough for a negative control swing (eg: 0 to -20dB), and on the right is a bipolar control swing (eg: +/-6dB). the output signal of the amplifier was measured for 1st, 2nd, and 3rd harmonics, and these were compared to the 0dBu output signal of the amplifier. the OTA generally didnt have much past 2nd harmonic, and the 2164 had components throughout the audio range, but i stopped at 3rd to save time, and because they were small enough to probably not matter. the test signal was a 1kHz tone.

i tested for both a 10k mode resistor and no mode resistor, and both with and without a linearization circuit (with a second 2164 in the feedback loop of an opamp). these are all done both with and without bleedthrough trimming on the amps.

for the bipolar inputs, the OTA performs really bad, as +10dB is a bit much and +20dB is well outside of its range (the base currents get rather large in the diffpair). but, for negative swings, it peforms quite well, pretty much at the same level as the linearized 2164 with no mode resistor. as for the 2164, the no mode resistor condition is much better than the 10k, and linearization peforms a few dB better across the range, regardless of mode resistor or trimming. one thing to note, is that the untrimmed OTA and 10k 2164 are completely awful, and pretty much unusable, whereas the untrimmed, no mode, 2164 is not that bad, performing better than the trimmed, 10k, 2164.

also, contrary to what figure 13 in the SSM2164 datasheet might suggest, i did not find any null where there was no control bleedthrough for a particular mode resistor value. i also did not find appreciable increase in distortion for the 2164 run with a trimmed input, but i want to run those tests one more time to be certain.

[ADDITION: 1/6]

so i did a check of distortion and noise on the audio pathway for using the linearization technique of a 2164 cell in the feedback loop of an opamp, and it seems pretty similar for both parameters to the non-linearized setup.

i also checked the distortion on the control signal pathway for a few different input levels. these arent exhaustive tests by any means, but i think they are useful:



the 2164 and OTA had a DC voltage applied to the audio pathway equivalent of +16dBu and +2dBu for the 2164 (5V and 1V across a 30k) and +3dBu for the OTA (relative to the OTA-SHIFT graphs above - 5V across 100k input resistor). both setups were trimmed for minimum bleedthrough, and the 2164 was tested for both 10k and no-mode conditions. linear input signals were applied of 1V-10V (-20dB), 0.1V-10V (-40dB), and 4.75V to 6.25V (~2.5dB), where 10V is 0dB gain.

the distortion was measured across the first 8 harmonics. for the 2164, the harmonics spread across the full audio range, and did not drop that quickly. by the 8th harmonic they were down by 10dB to 20dB, so i stopped taking them. for the OTA, the harmonics rarely passed the 3rd before they were into the noisefloor. the OTA performed an order of magnitude better, as is to be expected, as its not going from linear, to log, and then back again. if the distortion was graphed on a linear amplitude scale (rather than dB), the OTA distortion would make a straight line.

thats the end of it for now. there are a few other details tucked into the thread mentioned at the top of this post (mostly relating to the internal currents of the 2164), so maybe its worth checking out if you came here first. the output noise versus input resistor graph there is pretty useful.
J3RK
Very cool! Thanks for posting your findings! I use most of these ICs in various ways, (just starting with the 2164) and this information (and the other thread) are great resources.

thumbs up

Still need to check out that patent doc that Neil posted too.
E.A.S.
guest wrote:

-- more to come --


I love these types of threads.

Keep up the good work!

It might be a little late, but, The Linear Tech LT1228 might be a good chip to add to this comparison.

LT1228 PDF

-Mike
guest
thanks for the reminder about the lt1228. ive looked at it before, and it has some interesting features, but lacks any linearization scheme, so distortion is pretty bad. the datasheet has an interesting trick for keeping gm constant over temperature, by using a "bandgap" reference to feed the input bias stage. the reason why bandgap is in quotes, is because they also suggest just using a voltage divider to create 2.5V with similar results.
plushterry
This is really good. Thanks thumbs up
J3RK
guest wrote:
thanks for the reminder about the lt1228. ive looked at it before, and it has some interesting features, but lacks any linearization scheme, so distortion is pretty bad. the datasheet has an interesting trick for keeping gm constant over temperature, by using a "bandgap" reference to feed the input bias stage. the reason why bandgap is in quotes, is because they also suggest just using a voltage divider to create 2.5V with similar results.


I haven't given it more than a passing glance, but it looked more geared toward high-speed / video applications. However, for the distortion, I wonder if a dual OTA pre-distortion scheme could be used?
guest
there is only one per package (maybe they make a dual?), so matching might not be so great. i had another look at their THD graphs, and they seem to match the LM13700 for linearizing diodes, so perhaps they do have something at the front end, but just dont show it on the layout? it seems odd to put in something like that at not talk about it/show it.
neil.johnson
Remember that the 2164 was originally designed for audio mixing desk automation, where the most critical operating point is 0dB (straight through), so it is perfectly reasonable for the 2164 to excel at this position, and compromised at gains outside of this -- the gain rarely goes above 0dB, and by the time you're down to -40dB or lower the main concern is noise floor.

On the matter of sliding bias, as I pointed out in the other thread (or it may have been on the synth-diy list which is also having this type of discussion):

http://bitsavers.trailing-edge.com/components/pmi/_dataBooks/1990_PMI_ SSM_Audio_Products_Volume_1.pdf

on page 16 the sliding bias generator is presented, and the notes specifically call out sampling the output signal for compensating high-gain applications.

Neil
neil.johnson
Given that in the previous thread you wrote:
guest wrote:
i fully understand how the 2164 works. i know what those feedbak paths do. if you would like i could explain it to prove it, but i think that would just be a waste of both our times. i think we just have a different view of what "very different means".

I am thoroughly enjoying reading this new thread thumbs up

Neil
neil.johnson
One question: how many devices are you testing? I don't know about the LM13700 but the 2164 has quite a wide range of device-to-device behaviour in terms of noise and distortion - check out figures 5 and 7 in the ADI datasheet. The curves in the infamous figure 6 are the average of 1200 channels.

Neil
guest
ive only tested 2 devices so far. they both performed similarly, but not exactly the same. comparing to the datasheet curves, both devices looked like they were in the ballpark of what was shown, so my data collected here isnt too far off from "average" (assuming the datasheet was showing some sort of "average"). i mostly want to get a feel for where the 2164 peforms best. its an obvious win for mixing desks, but it also gets used for a bunch of other things in the synth world: VCO, VCA, VCF, ring mod, etc. and for some of these applications the control gain needs to swing over a wide range. for example, when i did my comparison of exponential converters, i was really disappointed with the linearity of the 2164 in that application, and now i understand why: the distortion suffers as you move away from 0dB, which you necessarily have to do in an exponential converter. you could use it in a single sided mode (only negative gains), but then your temperature compensation suffers.
neil.johnson
guest wrote:
ive only tested 2 devices so far. they both performed similarly, but not exactly the same. comparing to the datasheet curves, both devices looked like they were in the ballpark of what was shown, so my data collected here isnt too far off from "average" (assuming the datasheet was showing some sort of "average"). i mostly want to get a feel for where the 2164 peforms best.

Some data I've seen the THD+N at 0dBu can be anywhere from 0.025% to over 0.3% - that's more than a 20dB spread.

Quote:
its an obvious win for mixing desks, but it also gets used for a bunch of other things in the synth world: VCO, VCA, VCF, ring mod, etc.

Indeed, but I don't believe it was developed for that market - too small at the time. Mixing desk automation, on the other hand, is a much larger market (think of a large SSL mixing desk and how many VCAs in that beast).

Quote:
and for some of these applications the control gain needs to swing over a wide range. for example, when i did my comparison of exponential converters, i was really disappointed with the linearity of the 2164 in that application, and now i understand why: the distortion suffers as you move away from 0dB, which you necessarily have to do in an exponential converter. you could use it in a single sided mode (only negative gains), but then your temperature compensation suffers.

Again, this was never the intended application. It is a rather neat trick thanks to the reclusive Mike Irwin (I still haven't managed to contact him yet...) but I seriously doubt it ever cropped up in product planning meetings.

Neil
guest
just a bump to note that i added the bleedthrough measurements. i have distortion and noise on the control port to do, and then i think ill be done.
smrl
Thanks for the writeup!

Would you consider doing a comparison of the SSI2164's when they're available?
neil.johnson
smrl wrote:
Thanks for the writeup!

Would you consider doing a comparison of the SSI2164's when they're available?

From the tests I've run and from the large datasets I've seen the SSI2164 is a few dB quieter than the SSM2164.

One issue with these devices is the spread in performance - easily of the order of 20dB or more - so to get any meaningful comparison you really do need to do the analysis across a statistically-significant number of devices, otherwise you can't draw any meaningful comparisons. Even just between different channels o the same device you can see measurable differences.

Neil
Graham Hinton
neil.johnson wrote:

Indeed, but I don't believe it was developed for that market - too small at the time. Mixing desk automation, on the other hand, is a much larger market (think of a large SSL mixing desk and how many VCAs in that beast).


The E and G series could have up to 65 or 130 VCAs per console, but they were dbx202c or dBx202x. SSM eyed that market enviously, but they couldn't deliver reliable products with good enough quality or decent quality control (the same problem they had with SCI). I suspect it was aimed at the low end retrofitted automation market that had a brief bubble around that time (stick a VCA in the insert jack of a cheap mixer).

They used to turn up and shower me with samples which turned out to be buggy and have wide spreads. The most disconcerting thing was that their response was always to give more samples from the same batch instead of actually addressing serious problems. That's the problem with small scale silicon manufacturing, you have to sell what you've made and can't just write a whole batch off after packaging so you have to blag it.

The classic one was the SSM2300 drop in replacement for the classic 4051 based sample and hold. If you had alternate outputs at maximum and minimum values the whole chip turned in to an octal sawtooth generator! Did they fix it? Did they hell. They stood in front of my scope and pretended it wasn't happening. I asked them if they would like to sit in a control room with a Black Sabbath mix while every fader ramped to +10dB at an audio rate... I've been very cautious about designing in anything with SSM on the front ever since.
guest
ok, so i think ive taken all the data im going to take. id just like to conclude with a few opinions on the relative merits of the 2164 in comparison to the OTA. first off, it has way more headroom, but for modular synth circuits im not sure thats a really big win. signals tend to be 0-5V or +/-5V, with rails maxing out at +/-12V, so a 6dB signal increase is already pushing saturation of the opamps. granted, you could run the 2164 at +6dB and still have the same distortion characterisics as the OTA, but with better SNR. so for VCA applications, the 2164 clearly has better noise and distortion than an OTA when used with a 10k mode resistor. also, anywhere there is need for exponential control, it is far less parts count, so VCF integrators seem like a win as well. but, when you start getting into linearizing it, the advantages arent as great. the audio pathway is still superior, but the bleedthrough and distortion on the CV input are much worse. you have to go with no mode resistor to get decent bleedthrough, but that gives you worse distortion on the audio pathway. the possibility of getting away without trimming is attractive, but im totally extrapolating here from the 1 unit i tested, so who knows? maybe i just got lucky. from my work with OTAs, they always need trimming. so, for things like ring modulators where there is audio rate modulated, linear VCAs, i think i would prefer the OTA. as for its use an exponential converter in VCOs, i might go back and revisit my work there, as i now understand the distortion mechanisms a lot better, and might be able to fix the issues i had with linearity (but maybe not - needing a small mode resistor for good linearity on the Vt compensator might counteract the need for high mode resistor for good linearity on the exponentiator). the 2164 temperature compensation technique is good for small temperature deviations (+/-20C), but not so good beyond that.
megaohm
Thanks, guest, and everyone contributing to these 2164 threads.
I don't have much to contribute but wanted to give a "greatly appreciated" to you guys for all the hard work and sharing of experience.
Really valuable stuff and it has kept me checking back here regularly.
Thanks!
we're not worthy

guest wrote:
ok, so i think ive taken all the data im going to take. id just like to conclude with a few opinions on the relative merits of the 2164 in comparison to the OTA. first off, it has way more headroom, but for modular synth circuits im not sure thats a really big win. signals tend to be 0-5V or +/-5V, with rails maxing out at +/-12V, so a 6dB signal increase is already pushing saturation of the opamps. granted, you could run the 2164 at +6dB and still have the same distortion characterisics as the OTA, but with better SNR. so for VCA applications, the 2164 clearly has better noise and distortion than an OTA when used with a 10k mode resistor. also, anywhere there is need for exponential control, it is far less parts count, so VCF integrators seem like a win as well. but, when you start getting into linearizing it, the advantages arent as great. the audio pathway is still superior, but the bleedthrough and distortion on the CV input are much worse. you have to go with no mode resistor to get decent bleedthrough, but that gives you worse distortion on the audio pathway. the possibility of getting away without trimming is attractive, but im totally extrapolating here from the 1 unit i tested, so who knows? maybe i just got lucky. from my work with OTAs, they always need trimming. so, for things like ring modulators where there is audio rate modulated, linear VCAs, i think i would prefer the OTA. as for its use an exponential converter in VCOs, i might go back and revisit my work there, as i now understand the distortion mechanisms a lot better, and might be able to fix the issues i had with linearity (but maybe not - needing a small mode resistor for good linearity on the Vt compensator might counteract the need for high mode resistor for good linearity on the exponentiator). the 2164 temperature compensation technique is good for small temperature deviations (+/-20C), but not so good beyond that.
MUFF WIGGLER Forum Index -> Music Tech DIY  
Page 1 of 1
Powered by phpBB © phpBB Group