A LM13700-based VCA running on 3.3V (or: old wine in new bottles)

[jdbakker]

Dear all,

Several of you will know the LM13700 (second sourced as the NJM13700), a dual OTA (Operational Transconductance Amplifier). It is specified as having an (unipolar) supply range of 9.5 to 32V. For my purposes I needed to go quite a bit lower. Looking at the specifications and the functional block diagram (http://www.ti.com/lit/ds/symlink/lm13700.pdf p.9), it seemed that for low voltage use the Darlington buffer might be an unnecessary waste of headroom. Perhaps reducing the output compliance range and the input common-mode voltage could help too?

The attached schematic is the result of this exercise. Signal input is at the left side of C2 (~2Vpp max), control voltage enters at R1. D1-3, R8 and R7 bias the input to the middle of U1's common mode range. Improved Howland current pump U2 applies the CV as a current to the amplifier bias input. With the components as shown CV swing is from 0V (off) to 2V (full). R1/R2/R4/R5 are what I happened to have lying around; for best performance R1 should be a smidge higher (see TI AN-1515 for details; originally by Robert Pease but I see they left his name off in the latest version). The transimpedance amplifier around the other half of U2 plus C5, C6 and R9-11 reduce the swing on U1's output to 'almost' zero (as stiff as a virtual ground on a 1MHz op-amp gets, anyway). C6 could be reduced by at least a factor of 10, but for my purposes stability on a rickety breadboard was more important than audio bandwidth. While way outside the manufacturer's recommended operating range, the single sample I breadboarded appeared to work within spec (as far as I could be bothered to test).

[But why? This is for step two of an Introduction to Electronics-class. The central theme is a low-voltage modular synthesizer, with a VCO+ADSR in a microcontroller, but full analog VCF, VCA, log/antilog, rectifier, peak detector, discrete sequencer and much, much more. In step one the freshmen got to use my pre-made PCBs, now they get to Eagle up their own. The entire system runs on 3.3V because microcontroller, so I needed an inexpensive low-voltage VCA plan for them to use. Of all the contenders, the LM/NJM13700 is still surprisingly easy to get and comes in breadboard-friendly DIP]

Thoughts?

JDB.

 

[JohnRoberts]

I used the 13600 to make a voltage controlled sine wave generator in my old TS-1 back in the 80s.

I didn't try to run it from only 3.3V.

Looking at the functional schematic looks like input bias at 1.6V is borderline to keep the amp bias current mirror out of saturation (and that schematic may not be complete). 

Perhaps run it on your prototype from a variable PS and see how low before it stops working, and if it works differently up at 4V or 5V vs 3.3V? (cold temps are probably worst case).

JR

 

[jdbakker]

Looking at the functional schematic looks like input bias at 1.6V is borderline to keep the amp bias current mirror out of saturation (and that schematic may not be complete).

I bias the input at three diode drops + the drop over R8; at room temperature in my test setup that works out to almost exactly 2V. It is the output that is biased at Vcc/2, or approx 1.6V.

Perhaps run it on your prototype from a variable PS and see how low before it stops working, and if it works differently up at 4V or 5V vs 3.3V? (cold temps are probably worst case).

Interesting question indeed, took me a while to get to it with the holidays and all.

It was a bit tricky to decide what exactly to control with the variable supply. I decided to keep the current through the linearizing diodes (set by R7) constant, and to feed U2 with a fixed 5V so that the current source would not run out of headroom. I attached the variable supply to V+ of the LM13700 and to the top of R9, to keep the output biased at mid-supply. Tests were done with a 1kHz@2Vpp input. R8 was trimmed for minimal distortion; the controlling current source (Howland current pump) was set for an output of 2Vpp.

At 3.3V the THD was 1.6%, predominantly 2nd harmonic*. Increasing the voltage to 5V made no appreciable difference. Reducing the supply voltage below 3V started showing an increase in THD, with 3rd harmonic dominant. Interestingly the output level was dropping at this time (with no obvious sign of clipping).

I also tried reducing the input bias voltage by shorting out D3. This reduced the input bias to 1.56V. At 3.3V the output level was down by 25%; THD slightly higher than in the first test (1.75%). This setup worked at lower supply voltages, with THD abruptly rising around 2.5V.

I've no cooler spray or ice box at hand here, so no low temperature tests yet.

*I hadn't expected the second harmonic to be as dominant as it was, about 20dB higher than 3rd even after trimming R8. One of these day's I'll try differential drive on the input, to see if that helps.

JDB.

 

[JohnRoberts]

As I recall in OTAs the trade off for distortion was S/N... reducing the Vp-p in the input stage reduces distortion but degrades  noise performance.

JR

 

[jdbakker]

As I recall in OTAs the trade off for distortion was S/N... reducing the Vp-p in the input stage reduces distortion but degrades  noise performance.

Yes, and the data sheet even has graphs on how much distortion to expect for a given swing (figure 13, mostly agreeing with my measurements). If I drop the input level, the distortion goes down as expected.

What is still unclear to me is the reason for the significant 2nd order distortion in what is mostly a balanced design (inside the 13700, that is), and why the transconductance (/output level) drops with reduced input bias voltage. Could be the current source at the bottom of the LTP going outside its compliance range, or maybe the drop across 1k resistor I use to bias the linearizing diodes (R7 in my schematic) isn't large enough to make the dV at that point insignificant.

Enough to ponder.

JDB.
[and maybe I should do such pondering earlier than 2AM, might help in seeing things more clearly...]

 

[JohnRoberts]

The linearizing diodes (Gilbert?) could work better driven from a current source, but that may add too much complexity.

JR