(Diode) Bias Tremolo

[Rob Strand]

I am aware of the limitations of using diodes like this. Frankly, I'm surprised how well my circuit worked straight off the drawing board.
The diff amp can (almost) deal with a lot of gross behaviour from the diodes.
Also, the design goal was to design a tremolo that had inherent distortion at the heart of the tremolo effect. The non-linearity of the diodes is very much exploited to achieve this!

I thought it turned out really well too - and I'm across all the evils of diodes and stuff the RG was talking about.
Some of the diode bridge circuits don't actually cancel the LFO as well as yours.  They don't enforce a good balance on the DC level so some of the LFO gets through.

[PRR]

Prior Art: CBS had a line of (tank and) broadcast level controllers (AGC/Limiter) which used a diode bridge between transformers.

Around 1960, transistor radios often had a B-C diode which was open for small signals and biased on when AVC action happened. But this is different because AM modulation (between tuned circuits) accepts huge 2nd order distortion without loss of final fidelity (such as it was).

The optional input diodes on LM3700 are another diode attenuator.

10mv-30mV per diode is a ceiling. More diodes gives a higher ceiling but also higher impedance and thus higher noise resistance at idle. On paper, 100 diodes series gives 40dB more signal for 20dB more hiss. That's a lot of soldering. Fairly simple buffered mutipliers can give the same effect in a transistor, two Rs, and a couple opamps per "100 diode". This does lead to the need for >60V in hand, so is not pedal-friendly. But for studio it is less frightening/costly than 300V supplies for a tube-VCA of lesser S/N.

[samhay]

>I thought it turned out really well too

Thanks

>Prior art.

Is was looking at the transformer designs the other day. This could be an interesting diversion for version 2 or 3

>Fairly simple buffered mutipliers can give the same effect in a transistor, two Rs, and a couple opamps per "100 diode"

I did briefly consider using rubber diodes to make tuning forward voltage an option.
I think this is a different plan though?

[PRR]

> using rubber diodes

The common rubber diode is "diode-like" only over a *narrow* range of currents. Especially at high multiplication. As a class A driver for a push-pull output, this may be good-enough. In my sample (Q5), it is a 10K resistor below 0.8mA and approaches 10K/hFE by 8mA.

This can be averted with buffering. A full solution wants 3 buffers per transistor. In many cases at least one is not needed.

The diode-line can, as you say, be trimmed with resistor value. For demonstration, I paired a 2N2222 with a 2N3904. At 1mA there was a dozen+mV difference, I trimmed that out. The overlaid curves track over 6 decades (120dB!) with a slight slope. I suspect the slope is incomplete modeling. If it is real, some compensation can be had by changing the audio signal resistors.


The VCA form does the right thing. The 3rd buffer in each fake-diode is not needed. I used a simple driver and recovery which does what your real plan does. (The 1r mixer does not stress SPICE.) (I don't know what your 100r does, so I omitted it.)


Yes, we are replacing 10 or 100 diodes with 76 transistors. But the transistors are cheaper, less work, allow direct trim.

[stringsthings]

Just found this thread.  Very nice tremelo sounds!  Following this for future updates.

[samhay]

Was away from the bench for a while, but when I got back put a version 2 together that adds pre- and de-emphasis and removes filtering from around IC2B.

It works. Does it work better than version 1? I'm not sure yet...

Schematic and vero layout (click for larger version):

[Rob Strand]

It works. Does it work better than version 1? I'm not sure yet...

Thanks for the update.   I guess it's a good as it can be.
IIRC your guitar was potentially noisy.

[BetterOffShred]

V1 is on my build list.. maybe I'll wait and hear how this one sounds by comparison..  thanks for sharing man!

[samhay]

>Thanks for the update.   I guess it's a good as it can be.
No problem, and thanks for giving me a nudge to make improvements.

>IIRC your guitar was potentially noisy.
Good point.
I have spent some time with this and my bass, which is much quieter than the guitar I had been testing with.
In conclusion, it seems version 2 is quieter, particularly if you don't turn the bias (depth) up all the way. It can handle fairly hot input - one reason I shied away from pre-gain/emphasis in the first place - and thus:

>V1 is on my build list.. maybe I'll wait and hear how this one sounds by comparison..

I suggest you build version 2. I will try to get new sound clips up over the weekend, but if you are itching to get the soldering iron out, I don't think you will regret choosing V2 over V1.

[Rob Strand]

I have spent some time with this and my bass, which is much quieter than the guitar I had been testing with.
In conclusion, it seems version 2 is quieter, particularly if you don't turn the bias (depth) up all the way. It can handle fairly hot input - one reason I shied away from pre-gain/emphasis in the first place - and thus:

There's a bit of a trade off between susceptibility to overload and signal to noise improvement.
Your network starts ramping up at about 276Hz /278Hz  which is quite a low frequency.
I'd probably push that upto at least 400Hz.

If you used 10k + 1.2k + 33n on the input and 2k2  + 270R + 150n (allows for 100k output resistor) on the output it should reduce overload susceptibility *without* losing the s/n improvement you currently have.  With these values the network starts ramping up at about 431Hz/438Hz.

[If you really wanted to get fussy, not for the case when the Tremolo is drive by a guitar but more when it is driven by other effects, you can shave some noise off by using 2k2 + 270R + 150R for the input stage (maybe 0.4dB s/n improvement) and replace the 10n on the input with 47n or 100n (0.3dB s/n improvement); totaling 0.7dB.]

[samhay]

>There's a bit of a trade off between susceptibility to overload and signal to noise improvement...

Indeed. There is probably scope to tune the pre-/de-emphasis I added, as I didn't do this by ear.
I figured I would bring the corner frequency down as low as I dared and then back it off if it overloaded. It didn't, so I kept the V2 values.

BetterOffShred - if you build V2, might be worth playing with the values Rob suggested.

[samhay]

Here's a clip of version 2.
Recorded with a touch of compression before it, then straight into an audio interface.
Apologies for the uninspiring playing, but I was trying to show off the range of bias and speed settings and to (hopefully) show how quiet it now is.

https://www.dropbox.com/s/sikvri2ggk6qumh/Bias_Tremolo_V2.mp3

It was recorded with my relatively noisy tele and having now spent some time with it played into my amp, I am pretty sure that most/all of the noise I/you now hear is coming from the guitar rather than this circuit.
Version 2 is definitely an improvement and this is probably about as realistically quiet as a stompbox tremolo is going to get. I wouldn't have any qualms playing with it at stage volume.

[samhay]

I've had a question about how one might go about substituting another LFO into this circuit.

With respect to the V2 tremolo schematic:
IC1B is a phase shift LFO - like the EA tremolo and Magnavibe, but done with an op-amp. The LEDs (D4 and D5) limit the voltage swing of the LFO to about 3.5 V peak-to-peak, which gives a good effect and keeps IC1B from hitting the supply rails.

The output of IC1B (the LFO) swings around a bias voltage set by the BIAS control, which can vary between about 3-6V (with a 9V supply). The LFO doesn't drive anything, but rather acts as a virtual ground, which happens to wiggles up and down.

Now to the signal: D1 and D2 act as diodes that clip to ground - think Dist+ or Rat. However, instead of clipping to ground, they clip to the LFO virtual ground, which is wiggling. When the LFO wiggles down (towards ground) and/or when the BIAS is set higher (DC offset is closer to 3V), there is more clipping and vice versa.
There is also some region where these diodes are better thought of as variable resistors (more like a typical tremolo), but this region is quite small compared to the 3.5V LFO swing.
Also note there is a single 'clipping' diode for each signal path, so only one half of the signal gets clipped.

The reason this doesn't sound like a distortion is because the signal is phase split before it hits D1 and D2 and then recombined afterwards. The half of the signal that gets 'clipped' is opposite for the signal that comes out of IC1A and IC2A, so when the diff amp puts the signal back together again, it can use each half of the non-clipped signal.