A/DA flanger with minimal low pass filtering

[12Bass]

Hey Everyone!

Not long ago I built the SAD1024A version of the A/DA flanger clone.  Since then I've been tinkering with the design both to learn about the functions of each part of the circuit, and also to see if I can make some improvements. 

I know that one of the criticisms of the A/DA design is that its extremely wide delay range can cause the signal to become lost in the top part of the sweep.  So, as an experiment, I've removed the low pass filtering capacitors in the feedback loop (C18, C19) and also reduced the input filter on the BBD from 0.02 μF to 47 pF (C10, C34).  To illustrate how this has opened up the top end of the sweep, using an Echo Gina24 I've recorded a short sample using my Hamer 12-string bass which was patched through a VT Bass pedal in order to generate extra harmonic content. 

In order to better demonstrate the high frequencies produced, there's more feedback (Enhance) employed here than might be considered tasteful.  In other words, this recording isn't supposed to be an example of great tone so much as an experimental reference.  There has been no signal processing other than normalization and conversion to high-bitrate mp3.

A/DA flanger with minimal low pass filtering: http://members.shaw.ca/webplace/ADA with minimal lowpass filtering.mp3

As a result, rather than sweeping into a "black hole" up top, there is more harmonic content through the apex of the sweep.  Note that I've employed OPA1642 and OPA2211A op amps in the audio path (instead of LM324), as well as reduced the low pass filtering through the input and output stages so that there is more upper harmonic information left intact for comb filtering. 

[StephenGiles]

Interesting, the two chips that you used in the audio path are dual opamps of course so I would need to use a quad version to replace my MC 3403s. There is an OPA1644 but no quad version of the OPA2211A. What did you use where?

[12Bass]

BrownDog makes a dual SOIC to quad DIP adapter:  http://cimarrontechnology.com/2xdual-to-quadop-ampadapter.aspx

I used a somewhat more creative solution...



Going by the moosapotamus SAD1024A A/DA schematic, I used the OPA1642 for IC1a and IC1d, and OPA2211A for IC1c and IC1d plus IC2b and IC2c.

[Jarno]

  A testament to the high-temp performance of modern SMT components, nice one!

[StephenGiles]

No, I prefer to stick to the old fashioned size opamps

[12Bass]

No, I prefer to stick to the old fashioned size opamps

I generally do too; however, many of the new high performance parts are not available in DIP.  After a while, you get used to working with the little fellas.  From what I hear, the metal can versions tend to sound the best.  Haven't tried any yet....

One thing that I noticed with the reduced low pass filtering is that I'm hearing a bit of high frequency "garbage" when the feedback is turned up.  Not sure if it is aliasing from the clock, or   Whatever it is, the noise is minimized if I ground the Enhance pot (right now the circuit is loose and unshielded).  Given that I have some MN3007s on the way, looks like I'll be building a retrofit adapter board in the near future (thanks oldschoolanalog!).  Will be interesting to compare with the SAD1024A.

[12Bass]

After some more empirical research, I've recorded another sample, this time using a piece of music to demonstrate the flanger using a full-range signal.  A moderate amount of regeneration and a limited sweep was used for this recording.

http://members.shaw.ca/webplace/ADA with minimal lowpass music sample.wav

For the sample above, the A/DA clone's regeneration low-pass filter was removed (C18, C19 on input of IC1d).  This allows much greater regeneration at the top of the sweep range and allows the flanger to have more impact by comb filtering higher frequencies.  The original regeneration filter sounds smoother/mellower, as it filters out a lot of treble content.  Those interested in having a more dramatic sweep up high might find this modification worthwhile.  As it is, I didn't even sweep all the way up, as the frequencies affected at the top of the range are so high that it becomes difficult to hear the effect.  In fact, I find that the A/DA circuit's sweep range has to be limited somewhat as it can easily comb filter frequencies which are so high that they are barely audible.

As for the rest of the circuit, I've used a 22 pF capacitor for C8 (replacing 100 pF), 47 pF for c14 (replacing 100 pF in BBD reconstruction filter), 0.0047 μF replacing the original 0.01 μF C20 in the output filter, and 0.0033 μF replacing 0.02 μF (C10, C34) on the input of the SAD1024A.  The net effect is that the bandwidth of the flanger has been opened up for both straight and delayed signals.  This makes the flanger sound more transparent overall and also allows for a more dramatic flanging sound, especially in the higher frequencies.  I spent some time playing for various values on the input of the BBD and found that there was some audible clock/LFO interference with values smaller than 0.0033 μF, and that this value gave a reasonable compromise between bandwidth and noise.

From listening, my sense is that the gain of the BBD line various somewhat according to frequency.  The low to upper-midrange areas are more prominent and react more strongly to added regeneration (more "peaky" sounding.   Even with minimal low-pass filtering and considerable added feedback, the output drops off in the upper end of the sweep.  However, the modifications have made a significant improvement.

[Mark Hammer]

In most BBD-based circuits, the heavy filtering is what is needed to keep the noise and aliasing under control.  Of course, in many of those instances, the clock rate used is slow enough to present some seriously audible clock noise and aliasing.  The A/DA uses much less filtering, but uses a noise gate instead.  So I guess the key question here is "What does it sound like when you stop playing?".

[12Bass]

In most BBD-based circuits, the heavy filtering is what is needed to keep the noise and aliasing under control.  Of course, in many of those instances, the clock rate used is slow enough to present some seriously audible clock noise and aliasing.  The A/DA uses much less filtering, but uses a noise gate instead.  So I guess the key question here is "What does it sound like when you stop playing?".

Thanks for the comments, Mark!

Hmmm.... not overly noisy from what I can tell.  As I mentioned, I heard some clock/LFO noise with the low-pass capacitor on the BBD removed if regeneration was added.  However, using 0.0033 μF to ground seemed to reduce that noise to barely audible when not playing, and inaudible with a signal passing through.  Not sure if it is aliasing noise or something else.  Also, bear in mind that the circuit is presently not mounted in a shielded box.... just loose on the floor, so that's probably not helping matters.

From what I've gathered, the A/DA's lowest sampling rate is in the 69 kHz range (using parallel-multiplex), so that puts the Nyquist frequency over 30 kHz, where there shouldn't be much signal anyway.  So, I don't know if aliasing is much of a worry with this circuit.  I've also read that parallel-multiplex improves SNR by 6 dB.  Can anyone confirm this?

[12Bass]

Another sample with negative flange, more regeneration, and wider, deeper, sweep:

http://members.shaw.ca/webplace/ADA with minimal lowpass negative flange.wav

Notice how much louder it sounds at the low end of the sweep.

[StephenGiles]

Very interesting indeed, I'm on a hotel wifi at the moment in Marbella, Spain on the 3rd floor so it's difficult to listen to the samples all the way through. Looks like I'll have to dig out the breadboard when I return home!

[12Bass]

Here's a sample with a high-regeneration negative flange using the whole sweep range:

http://members.shaw.ca/webplace/White Noise negative flange.wav

I've halved the size of the high-pass capacitor which feeds the signal back into the BBD in order to reduce boominess.  As you can hear, it really gets way up into the high frequencies!

[Mark Hammer]

That's the miracle of the A/DA.  It just keeps sweeping up and up and up, into those regions where you didn't realize there was any more "up" up there to go to.

[12Bass]

Spent a little more time playing with the low pass filtering last night.  My goal has been to try to get the delay line to sound as clear and clean as possible, while avoiding obvious aliasing/sampling artifacts.  The thought behind this is that a clean, wide bandwidth, delay path will allow for the most dramatic flanging effect.  For testing purposes, I've been disconnecting the straight path and monitoring the output of the BBD line using K702 headphones.   

At the output of the BBD, there is a reconstruction filter in the feedback loop of of IC2c, namely C14.  Originally, it was 100 pF, however, I thought it sounded better/clearer with a 47 pF silver mica, so I've been using that instead.  Last night I tried using a 39 pF ceramic, but it sounded sort of metallic and coarse in the treble range.  I'm wondering if that might be the sound of the ceramic capacitor, or perhaps 39 pF just isn't enough to smooth over the sampling glitches in the output of the BBD?  Surprisingly, the isolated delay path actually sounds reasonably smooth and "hi-fi" with the relaxed filtering using the 47 pF silver mica for the post low-pass filter and 0.0022 μF at the input of the SAD1024A.  Some of that fidelity also seems to be due to the higher quality op amps; swapping LM324s into the audio path immediately sounds more one dimensional and "lo-fi".

Also did some more work calibrating the BBD bias setting.  The optimal bias definitely changes a little depending on the clock frequency.  After careful adjustment using a sine wave, I've found what seems to be a reasonable compromise where it distorts a little more at both frequency extremes, while remaining clean in the middle delay range.  There's definitely a drop in BBD output as the sampling frequency goes up, perhaps a few decibels near the highest point.  However, noise is lower and fidelity is noticeably better for higher sample rates.  Perhaps some sort of automatic gain adjustment might be a nice addition, to compensate for the signal loss at higher sampling rates.     

Overall, I'd say that the relaxed low pass filtering facilitates a more "transparent" and natural-sounding flange (or chorus).  Even without regeneration, you can hear the subtle reinforcement and cancellation as the delay sweeps up and down.  I'd suspect that the stock filtering would be considerably darker/murkier.  There might be a little more noise, but it hasn't been a concern.  The only thing that has been a bit of an issue is some high frequency "hash" when using large amounts of regeneration.  But that may be due in part to the fact that the circuit isn't properly shielded at present.     

Another thing I was wondering about is the diode limiter at the input.  From the old A/DA TZF thread, it was suggested that the limiter used in the Moosapotamus circuit was actually designed with the MN3010 in mind, a device with somewhat greater headroom.  How difficult would it be to modify the limiter to keep the signal more under control for somewhat lower headroom of the SAD1024A?

[12Bass]

Here's a new sample using a Radiohead clip, sweeping between roughly the middle of the range and near the top:

http://members.shaw.ca/webplace/ADA_Flanger_July_31.mp3

[12Bass]

I'm hoping that this thread may be of some use to other effects building tinkerers....

After some more experimentation, I've found that there is definitely an issue with aliasing byproducts using the current relaxed low pass filtering.  If the flanger gets a 19 kHz sine wave signal, there are sum and difference products which are clearly audible at the output, even though I can't hear the fundamental.  However, I can't hear the aliasing artifacts when running a full-range music signal.  These byproducts are only audible at the low end of the delay range, where the signal is closer to the clock rate.  

My suspicion is that, with a musical signal, a lot of the higher frequencies which cause aliasing are not sustained, but transients, or overtones which are at a relatively low level.  So, the net result is that aliasing noise is an issue for high frequency sine waves, but, practically speaking, not for music, and even less so for guitar/bass.  On the other hand, the relaxed filtering opens up the sound of the flanger, which makes it sound better IMO.  

Also did some testing with my Boss DD-20 and found that the A/DA's low point (14 ms) creates a resonance at roughly C#, so that might be of assistance for those wanting to calibrate their flangers and don't have access to a frequency counter.

From measurement, I found that the output of the BBD varies significantly depending on clock rate.  The output is greatest at lower rates and drops off as the rate goes up.  It doesn't sound quite like a linear function, as there is a greater drop at the very top of the range.  This variation in output makes it a challenge to achieve an optimal 50/50 mix between straight and delayed signals, as the levels only match precisely at one point during the sweep.  This appears to be due to a few factors, including the insertion loss of the BBD and the low pass filtering.  

Optimal bias also varies, with (IIRC) a slightly lower voltage being best at high clock speeds, and a higher voltage at low clock speeds (the variation is ~ 30 mV).  This requires setting the bias at a compromise which balances between the two.  Current demand also increases significantly as the clock rate goes up.  However, the L7815 appears to be able to keep the voltage constant.  I've used a bit more local bypassing on the clock and BBD which might also help.  

Background hiss is greatest when using long delays, and barely audible once in the middle range.  From what I can tell, the diode limiter circuit is insufficient to stop the BBD from overloading, at least with an SAD1024A.  Perhaps a somewhat lower limiting threshold might be helpful?  Not sure how the input limiter works with the MN3007 version, which should have a little more headroom.

In any case, I've definitely learned a lot in the process of building and experimenting with this flanger!  Thanks go out to all those who have helped along the way.

   

[StephenGiles]

It seems that you have given this a lot of time and thought, but I just wonder if any subtle changes to the circuit would be heard over a loud drummer and bass player (trying to play second lead guitar!!!)?

[12Bass]

It seems that you have given this a lot of time and thought, but I just wonder if any subtle changes to the circuit would be heard over a loud drummer and bass player (trying to play second lead guitar!!!)?

Good question.... 

How audible would the modifications be in a live situation?  The net result of the modifications is clearly audible when listening to just one source.  The increased high frequency response and improved signal path may help the flanger cut through the mix a bit better.

To help explain, my interest in building a flanger goes back many years to my youth when I built the Archer Electronic Reverb project using an SAD1024A.  That particular circuit was not a very good design, IMO, as it was noisy, made a crappy reverb sound, and had clock speeds which could only get into the chorusing range.  So, years later, I started wondering if I could put that old Reticon BBD to better use.  The A/DA flanger seemed to be the best project choice.

As my nature is to explore and refine, I've been experimenting with alterations to the design to see what improvements are possible.  I've spent a lot of time going through old threads, reading posts by yourself, Mark Hammer, oldschoolanalog, and others, trying to learn what I can about various designs, technological compromises, and solutions.   For me, it is not so much about the end result as the process of learning along the way.  Then again, I should probably focus on my thesis instead....