v2164 linear VCA problem

From circuitbending to homebrew stompboxes & synths, keep the DIY spirit alive!

Moderators: Kent, luketeaford, Joe.

Post Reply
User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1692
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Post by neil.johnson » Fri Dec 15, 2017 5:47 pm

Hmmm.... my little paralleled VCA circuit seems to behave itself - probably due to the different input configuration compared to the standard 30k resistor. The only explanation for the distortion at high amplitude I can see is the input buffer op-amp distorting, as I see the uptick even at high levels of attenuation (so it's not the output path).

I'll need to dig through my notes to see what I saw before. You certainly need the input compensator - without it the 2164 is rather jumpy!

Also, the 2164 is a rather widely varying device - while the curves I see are the same shape as your measurements, the positions are different. But from what I've seen this is not surprising.

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

User avatar
Dr. Sketch-n-Etch
Super Deluxe Wiggler
Posts: 8114
Joined: Thu Jan 21, 2010 1:46 pm
Location: Vancouver, BC, Canada

Post by Dr. Sketch-n-Etch » Sat Dec 16, 2017 3:44 am

guest wrote:here is my noise analysis for varying input resistance:

Image

this was measured with a 30k output resistor. the low frequency "noise" is actually oscillations and amplification of line noise. i had to clean up my setup to get those values down to that level. the high frequency noise doesnt seem to have any structure.
Well, I think this answers my question. I built a crossfader circuit with 100k resistors, and it worked swimmingly and sounded great. I'm now going to build all of my 2164 circuits with 100k, and not think twice about sending parallel inputs into the VCAs when the need arises.
Power spent passion bespoils our soul receiver
Surely we know.

User avatar
guest
Super Deluxe Wiggler
Posts: 5852
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Sat Dec 16, 2017 5:18 am

neil.johnson wrote:The only explanation for the distortion at high amplitude I can see is the input buffer op-amp distorting, as I see the uptick even at high levels of attenuation (so it's not the output path).
yeah, this is my thought as well, although im not certain the mechanism. i was expecting it to just be the lack of gain such that the summing node isnt really kept at 0V, but the numbers arent working out for that.
neil.johnson wrote:Also, the 2164 is a rather widely varying device - while the curves I see are the same shape as your measurements, the positions are different. But from what I've seen this is not surprising.
i also made a mistake in my calculations for that graph. all the lines should move up 3dB (sqrt2 error). but variation accounts for a lot. the OTA lines are the right level, though, and i fixed the distortion, so its a straight line down to its noisefloor, rather than that wierd curved shape.
openmusiclabs.com

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1692
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Post by neil.johnson » Sat Dec 16, 2017 6:38 am

guest wrote:here is my noise analysis for varying input resistance:

Image

this was measured with a 30k output resistor. the low frequency "noise" is actually oscillations and amplification of line noise. i had to clean up my setup to get those values down to that level. the high frequency noise doesnt seem to have any structure.
In your analysis do you account for the increase in thermal noise for the input resistor?

Also, what is your vertical axis relative to? Your plot says "dB", but I suspect you mean "dBu" or perhaps "dBV"?? Otherwise you need to specify a reference level.

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

User avatar
guest
Super Deluxe Wiggler
Posts: 5852
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Sat Dec 16, 2017 6:43 am

y-axis is dBv, and i didnt subtract the thermal noise of the resistor. i just swapped out resistors and measured the output noise.
openmusiclabs.com

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1692
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Post by neil.johnson » Sat Dec 16, 2017 9:18 am

guest wrote:im using an HP3561A.
Nice! I used one years ago when designing some high power audio filters.
What do you use for noise measurements?

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

User avatar
guest
Super Deluxe Wiggler
Posts: 5852
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Sat Dec 16, 2017 10:31 am

yeah, i really like that spectrum analyzer. its quite bulky and noisey by todays standards, but i like the user interface. the display gives off a 19kHz hum that sometimes gets picked up in my circuits, though. i read a great article on the design of it, and how they worked around the limited technology they had at the time.

for noise, i mostly just use the spectrum analyzer. the noise floor is extremely flat on most of the things im testing, so its an easy integration. otherwise i build a filter and send it to my benchtop multimeter. ive compared the 2 and they are pretty close. every so often ill break down and use a soundcard and a spectrum analyzer program on the computer.

on an unrelated note, i realized that the input distortion curves i made suggest that the input stage is a darlington, and not just a single diffpair (so the hoshuyama spice model is correct). a darlington input has half the gain of the regular diffpair OTA, which would give -43dB first harmonic for our values.
openmusiclabs.com

v8pete
Common Wiggler
Posts: 90
Joined: Thu Jul 18, 2013 5:32 pm

Post by v8pete » Sat Dec 16, 2017 11:24 am

Great thread this! It would be interesting perhaps to try a resistor from the MODE pin to ground, to shunt some of the MODE current; I tried 10k to Ground in the LTSpice model, and this halves the current in the two diff pairs (taking the total current from about 30uA to 16uA). This drops the noise a lot further, and everything still seems to work normally (in LTSpice)- I haven't measured the distortion in the model as this is a bit fiddly to do accurately. but will have a look later...

Have been on the lookout for an HP 35665A for a while, but the few that appear on Ebay are a bit pricey given their age - it was one of my favourite pieces of equipment at work.

User avatar
guest
Super Deluxe Wiggler
Posts: 5852
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Sat Dec 16, 2017 1:12 pm

so im now very convinced that the distortion in that +20dBu range is due to the finite gain of the input OTA stage (and therefore a small, distorted signal being developed at the summing node). i created a +/-333uA, sinusoidal current source with -65dB 2nd and 3rd harmonics and fed it into that node. the output had -65dB 2nd and 3rd harmonics. im still confused, however, as to why the other data doesnt show this. perhaps im just looking at it wrong.

as for reducing the mode current, the distortion will go up. you can probably guestimate how much if you look at the distortion vs mode resistor graphs. the "no mode" is actually 80uA, and the remainder are ~14.3V/Rmode + 80uA

EDIT: just redid my tests, and got the current source down to -75 2nd and -84 3rd (basically the level of my voltage source), and im getting the same results out of the 2164 with the current source drive as with the voltage source drive. although thats disappointing because it means i dont understand whats causing the distortion, at least its consistent with my other findings, so something makes sense. im now suspecting finite emitter resistance degrading linearity in some way. im not sure exactly where, though.

also, with regards to crossover distortion being the dominant mode in the below 0dB section, doesnt that usually increase with decreasing amplitude?
openmusiclabs.com

User avatar
guest
Super Deluxe Wiggler
Posts: 5852
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Sun Dec 17, 2017 4:46 am

i ran some test at 0dBu input and +20dB gain; and +20dBu input and -20dB gain. both of these would have the OTA input stage driving 300uA into one side the diffpairs. for the +20dB gain scenario, the output side gets the 300uA, whereas the input side gets 30uA. for the -20dB scenario, the input side gets 300uA and the output side gets 30uA. this is different from the +20dBu with 0dB gain tests i have done above, for which both sides get 300uA.

the distortion was much worse for the +20dB and -20dB gain scenarios, as compared to 0dB gain. and, the distrtion was identical for the two +/-20dB conditions. this leads me to believe that emitter resistance and finite base currents (across those 500ohm base resistors) are having a large effect as you deviate from 0dB gain. but, at 0dB gain, both of those effects should cancel each other. so im left with the current mirrors at the top of the diffpairs as culprits. perhaps beta degradation with higher currents leads to a mismatch btween the drive/recieve side of the current mirrors. since which side has the high current depends upon the polarity of the input signal, you would get a lot of second harmonic distortion. i will have a look at the distortion risiduals, which should give some clues. it could be that a minor version of this is whats causing the crossover distortion, as both smaller signals and higher mode current would both reduce the issue. interestingly enough, the THAT chips do away with this possible problem by not having current mirrors, and just using a PNP diffpair for the other side.
openmusiclabs.com

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1692
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Post by neil.johnson » Sun Dec 17, 2017 6:51 am

guest wrote:i ran some test at 0dBu input and +20dB gain; and +20dBu input and -20dB gain. both of these would have the OTA input stage driving 300uA into one side the diffpairs. for the +20dB gain scenario, the output side gets the 300uA, whereas the input side gets 30uA. for the -20dB scenario, the input side gets 300uA and the output side gets 30uA. this is different from the +20dBu with 0dB gain tests i have done above, for which both sides get 300uA.

the distortion was much worse for the +20dB and -20dB gain scenarios, as compared to 0dB gain. and, the distrtion was identical for the two +/-20dB conditions. this leads me to believe that emitter resistance and finite base currents (across those 500ohm base resistors) are having a large effect as you deviate from 0dB gain. but, at 0dB gain, both of those effects should cancel each other. so im left with the current mirrors at the top of the diffpairs as culprits. perhaps beta degradation with higher currents leads to a mismatch btween the drive/recieve side of the current mirrors. since which side has the high current depends upon the polarity of the input signal, you would get a lot of second harmonic distortion. i will have a look at the distortion risiduals, which should give some clues. it could be that a minor version of this is whats causing the crossover distortion, as both smaller signals and higher mode current would both reduce the issue. interestingly enough, the THAT chips do away with this possible problem by not having current mirrors, and just using a PNP diffpair for the other side.
Minimal distortion at 0dB gain is a good place to be for an audio VCA, since that is where is will spend most of its time, maybe a few dB up or down. Most mixing desk faders rarely go above +10dB gain, and by the time you get down to -20dB to -30dB that channel will likely be drowned out by other channels in the mix so the distortion products, while not desirable, will more than likely be of less a concern. If you take this mentality (0dB is sacred) to the extreme you end up with BBC-like VCA inverse-fader circuits.

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

User avatar
guest
Super Deluxe Wiggler
Posts: 5852
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Mon Dec 18, 2017 10:48 am

so i had a look at the risiduals, which was pretty fun. you can watch it go through the various modes of distortion. on the low side the gain is slightly different for positive and negative going halfwaves. the gain charactersitcs are inverted as well, whereas one side slightly rounds off the wave, the other side peaks it up a bit. in the middle regime, the emitter resistance becomes clear, as exponential peaks begin to develop. the final phase is clipping. this seems to agree with previous hypothesis.

i also had a realization for one reason why the 2164 is so much lower noise than the 13700: the average output stage current is lower. in an OTA, the diffpairs and current mirrors are always run at your bias current (gain setting), and then as you turn up the input signal, you start to slosh that current around more to get a bigger signal at the output. but, with the 2164, you start with a really small current in those transistors, and then as you apply a signal, you increase the bias current to get a bigger output signal. this means the noise floor in the 2164 increases with signal level, whereas it doesnt change much at all for the 13700 (although it sucks pretty hard to start out).

i ran some quick tests, and the noise floor definitely comes up. at a +17dBv input its at -76dBv, whereas with no signal its at -96dBv.
openmusiclabs.com

User avatar
guest
Super Deluxe Wiggler
Posts: 5852
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Mon Dec 18, 2017 12:10 pm

and here is a graph of SNR for 10k mode, no mode, and an OTA.

Image

you can see the two different regimes for noise. there is the intrinsic noise of the mode current on the left side, and then the amplitude based noise, which is fixed per amplitude regardless of mode on the right.
openmusiclabs.com

v8pete
Common Wiggler
Posts: 90
Joined: Thu Jul 18, 2013 5:32 pm

Post by v8pete » Mon Dec 18, 2017 5:31 pm

Great insights there Guest. I ran up my LTSpice model of a 13700 style OTA this morning, and although as you say, its quite a bit noisier than the 2164 at high bias currents, it still looks pretty darn quiet to me at "zero" bias currents (ie. at max attenuation - with all the current mirrors running with pA currents) which is of course the way you might want a synth final VCA to behave. I wanted to compare this to the 2164 model, and also look at the attenuation at 10kHz (at max attenuation) before concluding that the 2164 is definitely better in these respects - have to do that another day as the 2164 model is on my other PC!

User avatar
guest
Super Deluxe Wiggler
Posts: 5852
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Mon Dec 18, 2017 6:32 pm

im glad youre verifying this in spice, it helps reassure me that its not just some artifacts in my setup. i ran some more tests today comparing the noise floor of the 2164 versus an OTA for various bias levels. it was interesting that the OTA and 2164 noise levels lined up almost exactly with a sqrt2 (10dB/decade) curve, but the OTA noise shot up about 100uA or so, and started becoming a linear curve (20dB/decade). the only thing i can think of that creates the linear noise is base resistance noise. so the 2164 must be nulling this out with the feedback. with a 200ohm base resistance, the transfer point for shot noise versus thermal noise is around 100uA, which is what i am seeing.
openmusiclabs.com

User avatar
guest
Super Deluxe Wiggler
Posts: 5852
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Tue Dec 19, 2017 5:03 pm

i just figured out the noise differnce issue; its as i suspected, the base resistance is nulled out by the feedback loop. the noise in theory should be even higher for the 2164 because of the added 500ohms of base resistance exterior to the transistors, but the feedback loop does its thing, and cancels it both for the input and and output stages, as they are connected in those diffpairs. pretty cool.

EDIT: actually, its not the feedback, i spoke to soon. its the fact that the bases are connected together to the same noise source, and the output currents are subtracted from one another. so the noise is just common mode signal, which is cancelled out. although that isnt true for the internal base resistances, so maybe im still confused.
openmusiclabs.com

User avatar
neil.johnson
Super Deluxe Wiggler
Posts: 1692
Joined: Thu Apr 08, 2010 3:51 pm
Location: Cambridge, UK
Contact:

Post by neil.johnson » Wed Dec 27, 2017 7:42 pm

For those interested in the history of the Frey cell here is the original patent:
http://www.google.com/patents/US4471320

The second patent - describing a monolithic embodiment - is this one:
http://www.google.com/patents/US4560947

Neil
Random ramblings :: http://www.njohnson.co.uk
SSI parts and more :: https://www.cesyg.co.uk

User avatar
guest
Super Deluxe Wiggler
Posts: 5852
Joined: Mon Aug 19, 2013 11:49 am

Post by guest » Wed Dec 27, 2017 8:14 pm

thanks for posting that. i was just fixing a bunch of serge modules from 1979 this past weekend, and they were chocked full of figure 4 and 5 from his patent. i was wondering about the history of the gain cell.
openmusiclabs.com

User avatar
J3RK
Super Deluxe Wiggler
Posts: 6733
Joined: Tue Dec 22, 2009 12:46 pm
Location: Seattle

Post by J3RK » Thu Dec 28, 2017 12:47 am

Indeed that's very similar to the Serge cells. I've built a ton of those little sandwich cells :hihi: still have a small pile of them sitting on one of the benches. For some reason I thought the Serge ones were Blackmer(esque), but I haven't looked at the schematics in a LONG time, and the last couple of weeks is the first I've really been introduced to the Frey.

User avatar
otherunicorn
Santa
Posts: 528
Joined: Fri Feb 20, 2009 4:38 pm

Post by otherunicorn » Thu Dec 28, 2017 1:17 am

J3RK wrote:Indeed that's very similar to the Serge cells. I've built a ton of those little sandwich cells :hihi: still have a small pile of them sitting on one of the benches. For some reason I thought the Serge ones were Blackmer(esque), but I haven't looked at the schematics in a LONG time, and the last couple of weeks is the first I've really been introduced to the Frey.
I was told that Serge and Blackmer cells are very similar - as they were the result of the two people in question having a discussion.
You got that resistor in backwards...

User avatar
cygmu
Super Deluxe Wiggler
Posts: 1677
Joined: Mon Feb 15, 2016 3:33 pm
Location: UK

Post by cygmu » Thu Dec 28, 2017 2:32 am

And the Serge gain cells were in use well before the filing date of the Frey patent (1981) weren't they? Well, this is is interesting in a number of ways.

EDIT: Here's the Blackmer patent. http://google.com/patents/US4403199
I haven't read it but the figures are all (at least) 8 transistor designs, whereas as guest points out the Frey patent has a couple of four-transistor setups in its figures which do look more like the Serge cells.

User avatar
KSS
Super Deluxe Wiggler
Posts: 5298
Joined: Mon Jan 25, 2016 7:28 am

Post by KSS » Thu Dec 28, 2017 6:43 am

guest wrote:im glad youre verifying this in spice, it helps reassure me that its not just some artifacts in my setup.
This is pure gold. I'm sure Bob Pease and Jim williams just rolled in their graves.
Laughing. :lol:

v8pete
Common Wiggler
Posts: 90
Joined: Thu Jul 18, 2013 5:32 pm

Post by v8pete » Thu Dec 28, 2017 8:10 am

KSS wrote:
guest wrote:im glad youre verifying this in spice, it helps reassure me that its not just some artifacts in my setup.
This is pure gold. I'm sure Bob Pease and Jim williams just rolled in their graves.
Laughing. :lol:
Well, actually Bob Pease was quite ok with Spice; he mentions its use in a positive sense in quite a few of his written publications. I'm not sure though how often Jim Williams made use of spice, if ever - I don't recall reading him mention it much, and I've read a great deal of his material over the years. These two guys though had such an exceptional insight into analogue design that I doubt that they'd need to use Spice to figure out fundamental circuit behaviors,whereas for the rest of us it is an invaluable tool to use to poke around and see what's going on. LTSpice is of course used by the in-house chip design engineers at Linear Technology for IC design, so I guess that they're pretty happy with its behavior and accuracy these days - I certainly am! Of course, Bob Pease was particularity vocal about people who don't build stuff but just rely on simulation -I'm sure that he'd have no trouble at all with those of us who always try and do both, as they complement each other exceedingly well nowdays.

loki
Common Wiggler
Posts: 116
Joined: Fri Apr 25, 2014 4:20 pm

Post by loki » Thu Dec 28, 2017 1:07 pm

cygmu wrote:And the Serge gain cells were in use well before the filing date of the Frey patent (1981) weren't they? Well, this is is interesting in a number of ways.

EDIT: Here's the Blackmer patent. http://google.com/patents/US4403199
I haven't read it but the figures are all (at least) 8 transistor designs, whereas as guest points out the Frey patent has a couple of four-transistor setups in its figures which do look more like the Serge cells.
Blackmer's first VCA patent is 3714462 and was filed in 1971 along with the 3681618 which is an RMS circuit. These two patents were fundamental coverage of the noise reduction system for magnetic tape that dBx manufactured.

To go one step further back, look at 3518578 filed by Alan Oppenheim and Thomas Stockham. The same idea with diodes instead of transistors.

User avatar
cygmu
Super Deluxe Wiggler
Posts: 1677
Joined: Mon Feb 15, 2016 3:33 pm
Location: UK

Post by cygmu » Thu Dec 28, 2017 3:37 pm

loki wrote: Blackmer's first VCA patent is 3714462 and was filed in 1971 along with the 3681618 which is an RMS circuit. These two patents were fundamental coverage of the noise reduction system for magnetic tape that dBx manufactured.

To go one step further back, look at 3518578 filed by Alan Oppenheim and Thomas Stockham. The same idea with diodes instead of transistors.
Thanks! This history and development is fascinating and the patents are a pretty good way to learn what goes on, at least so it seems to me. If I had gone so far as to read the Blackmer patent I dug up I would have discovered this earlier one so thank you for filling in my laziness.

The earlier Blackmer patent does look a lot like the Serge gain cells as far as I understand them.

Post Reply

Return to “Music Tech DIY”