EM3207 (v1.1) - MN3207 based EHX Electric Mistress (9V) clone

[DrAlx]

 Removed post. Posted before completed typing....

[DrAlx]

OK I have now done a thorough test of the BL3207 vs SAD1024 in the V2 Electric Mistress Circuit.
Based on my measurements, my suggestion to drop Q1 and just use the default V2 circuit values works really well.
I measured less than 1dB difference between the two chips across almost all audio frequencies and clock frequencies in the range of interest.  There is one difference though that I will mention right at the end.

I made the following test setup to rule out any differences down to part tolerances.

1) I used a CD4047 chip to generate a reference clock.
    I tested with 4 clock frequencies that span most of the range used by the BL3207:  50 kHz, 100 kHz, 200kHz, 400 kHz.

2) The CD4047 drives a CD4050B clock buffer for the BL3207.  The 4050 gives a bit more gain than the CD4049B or CD4049UBE chips at high clock frequencies (I did tests for this before).

3) The buffered clock from the CD4050B also clocks a CD4013. The CD4013 therefore provides a clock for the SAD1024 that is exactly half the clock rate of the BL3207.

4) Both BBDs have their bias set independently and optimised for the highest clock frequency.
SAD1024 and its clock (CD4013) were powered from a 12V supply.
BL3207 and all other chips in the test were powered from a 9V supply.

5) Each BBD has its 2 outputs tied together, so there is effectively a single output test point on each BBD.

6) The test audio signal is a sine wave and the same signal goes to both BBD inputs.  I tried audio frequencies from 100 Hz to 20kHz.

7) I made a single test load based on the parts on the BBD output in the V2 EM.
I assumed the color trimmer has a fixed value of 3k9 and that the color pot was turned to minimum.
So the test load is this:

10k || 3n3 || (47n + (3k9 + 10k || [470 + 47n]) ||  (10k + [470 + 47n] || [13k + 47n]))

I built this as a standalone network.  For any given audio signal and clock frequency I just attach that load between ground and the BBD output I want to measure, and read the pk-pk voltage at the BBD output.  It's quick and easy to switch the load from one BBD to the other and get comparison of voltages.

I was shocked at how good the agreement was.  Following table shows BL3207 BBD gain relative to SAD1024 gain (dB) for the 4 test clock rates.  Negative values mean SAD1024 was stronger output.

Values are typically less than 1 dB different, showing that best way to get EM3207 to match original EM is to leave out Q1 completely and just use the same component values as the original EM.

Audio (Hz)   50kHz   100kHz   200kHz   400kHz 100   0.35   0.35   0.00   0.00 200   0.35   0.35   0.35   0.00 400   0.00   0.00   0.00   0.37 800   -0.41   0.00   0.00   0.00 1000   0.00   0.00   0.00   -0.45 2000   -0.51   -0.26   -0.26   0.00 4000   -0.58   0.55   -0.30   0.00 6000   -0.31   -0.56   -0.31   -0.65 8000   -0.31   -0.95   -0.31   -0.98 10000   0.00   -1.29   -0.67   -0.67 12000   0.00   -1.33   -0.67   -1.02 14000   0.00   -1.33   -1.07   -1.07 16000   -0.69   -0.37   -1.07   -1.07 18000   -0.76   -0.37   -1.11   -1.51 20000   -1.11   -1.07   -1.51   -1.16

EDIT:  When I first posted the table I got some signs wrong.  Have corrected it now.

So what's the big difference between the BBDs if the gains match so well?
Well based on my experience of taking the measurements using a DSO, when the clock rates get low and the audio frequency gets high, the output waveform on the BL3207 looks much nicer than the one from the SAD1024.  e.g. the BL3207 output still resembles a sine wave while the SAD1024 is outputing a mess!!!  That made taking some of the SAD1024 measurements hard.
The messy waveform is partly down to the lower clock rate and lack of balance trimmer, but I think there might be something else going on.  I noticed that when the BL3207 output eventually starts to distort, the output sine wave looked kind of triangular (i.e. pointy tops and bottoms, and you could still sort of see the sine wave).
When the SAD1024 output started to distort, the output wave looked kind of square (i.e. "flat tops and bottoms").  Now I know that both the triangle and square harmonic series consist of odd harmonics but in different ratios, so maybe the difference in these chips is in the distortion.  I wonder if the "clarity" people are hearing in the SAD1024 is actually down to the harmonic content coming from the distorted waveform?

[rankot]

Did you have time to adjust schematic to your findings and do a hearing comparison with both?

[DrAlx]

No, and the whole point of going through the above process and getting the numbers was to avoid a subjective comparison.

The EM3207 still has other differences in the audio path (e.g. input section with gain boost), and the LFO sweeps differently to the V2 (range is different and waveform is rounded off so it less triangular), making it hard to set up two circuits the exact same way.

EDIT: If you make these changes to Thomeeque's schematic

• Change (R12, R13, R14, C7) from (4k7, 4k7, 47k, 680nF) to (470, 470, 10k, 3n3).
• Remove Q1 and R15 completely.
• Connect the top of C7 directly into C8.

the result will be similar to V6 EM, but with boost at input.

http://www.metzgerralf.de/elekt/stomp/mistress/images/1981-electric-mistress-v6-schematic.gif

[rankot]

No, and the whole point of going through the above process and getting the numbers was to avoid a subjective comparison.

The EM3207 still has other differences in the audio path (e.g. input section with gain boost), and the LFO sweeps differently to the V2 (range is different and waveform is rounded off so it less triangular), making it hard to set up two circuits the exact same way.

EDIT: If you make these changes to Thomeeque's schematic

• Change (R12, R13, R14, C7) from (4k7, 4k7, 47k, 680nF) to (470, 470, 10k, 3n3).
• Remove Q1 and R15 completely.
• Connect the top of C7 directly into C8.

the result will be similar to V6 EM, but with boost at input.

http://www.metzgerralf.de/elekt/stomp/mistress/images/1981-electric-mistress-v6-schematic.gif

Great, thanks! I will certainly try that. I have bought a PCB for 3207 version from another board member, and was just thinking to populate it few days ago, but decided to wait when I saw this post. 

[PRR]

> Posted before completed typing....

The Modify button below (and above) your post allow you to finish-off an incomplete post.

There is some time-limit, but usually long enough to correct an "ooops".

[12Bass]

So what's the big difference between the BBDs if the gains match so well?
Well based on my experience of taking the measurements using a DSO, when the clock rates get low and the audio frequency gets high, the output waveform on the BL3207 looks much nicer than the one from the SAD1024.  e.g. the BL3207 output still resembles a sine wave while the SAD1024 is outputing a mess!!!  That made taking some of the SAD1024 measurements hard.
The messy waveform is partly down to the lower clock rate and lack of balance trimmer, but I think there might be something else going on.  I noticed that when the BL3207 output eventually starts to distort, the output sine wave looked kind of triangular (i.e. pointy tops and bottoms, and you could still sort of see the sine wave).
When the SAD1024 output started to distort, the output wave looked kind of square (i.e. "flat tops and bottoms").  Now I know that both the triangle and square harmonic series consist of odd harmonics but in different ratios, so maybe the difference in these chips is in the distortion.  I wonder if the "clarity" people are hearing in the SAD1024 is actually down to the harmonic content coming from the distorted waveform?

Is it possible that the SAD1024 exhibits a greater optimal bias voltage shift in order to retain a low distortion output?  I noticed that optimal bias changed somewhat depending on frequency with my SAD1024 A/DA flanger.  I set it to be best at the "sweet spot" in the sweep, which is somewhere around 1ms IIRC (not sure of the clock frequency).  I definitely noticed the gain went down a bit at the highest clock rates.  But its delay path sounded surprisingly high fidelity (not distorted).  Perhaps try optimizing the bias for somewhere nearer the middle of the range.  Note that I use past tense because my friend's son knocked it off of my amp a few months ago and I think that the SAD1024 may have died as a result and I haven't had the heart to troubleshoot beyond confirming that the clean path still works and the delay path is dead.

[DrAlx]

The max clock in my test corresponds to 1.28 ms and I set bias for that.
You would not hear the distorted waveforms I was seeing as what you would call a distorted sound because only the highest audio frequencies (above several kHz) see the effect, and only at the lowest end of the sweep. So all additional harmonic content in the distortion would be above 10kHz.

[12Bass]

The max clock in my test corresponds to 1.28 ms and I set bias for that.
You would not hear the distorted waveforms I was seeing as what you would call a distorted sound because only the highest audio frequencies (above several kHz) see the effect, and only at the lowest end of the sweep. So all additional harmonic content in the distortion would be above 10kHz.

When I was listening, I used full range music with lots of high frequency content and listened with AKG K702s.  My flanger also has considerably relaxed filtering before and after the BBD which works very well (very open/clear sounding) at higher clock rates but aliasing becomes an issue as the delay time increases.  That's not much of a problem for me because my favorite flanging sounds are probably in the 0.5ms - 4ms range.  Delay path fidelity was surprisingly high (close to bypassed sound) at high clock rates.  15ms sounded more "grungy" and low-fi and the BBD gain was noticeably higher as well.  Didn't look at the waveform on a scope, however.  Would still suggest optimizing the SAD1024 bias for a lower clock rate and see how that affects the output waveform.

[DrAlx]

I repeated the comparison test but this time set bias levels for both BBDs with clock at lowest frequency (25kHz on SAD1024, 50kHz on BL3207).
Setting optimum bias level for lowest clock frequency meant I was starting to see clipped outputs on both BBDs when testing with the fastest clocks, so I had to lower the signal level when that happened to keep the output waves looking nice and measurable.

Another change I made to each BBD was to add a 1k trimpot between BBD outputs to minimise clock glitches.  I set those trimmers at the start of the test when the clock was at bottom of range.

End result (table below) was that I had even better agreement between chips than in the first test.
Not sure if that was because of trimmers or different bias set up.

Audio (Hz)	50kHz	100kHz	200kHz	400kHz
100	0.00	0.00	0.00	0.38
200	0.00	0.00	0.00	0.38
400	0.38	0.00	0.38	0.00
800	0.00	0.00	0.00	-0.45
1000	0.00	0.00	0.00	0.00
2000	0.00	0.00	0.00	-0.28
4000	0.00	0.00	-0.33	-0.43
6000	0.00	0.00	-0.35	-0.45
8000	0.00	0.00	-0.37	-0.71
10000	0.00	-0.09	-0.37	-0.74
12000	0.00	0.00	-0.39	-0.76
14000	0.00	0.00	-0.40	-0.78
16000	-0.40	0.00	-0.43	-0.80
18000	-1.34	0.00	-0.45	0.00
20000	-1.59	-1.27	-0.88	-0.86

So leaving out Q1 in this circuit gives a favourable comparison between chips (less than 1dB difference in wet path for most frequencies) at both ends of the bias range.  Probably is good for all other bias levels in between too.
Most people can't hear a 1 dB change in volume (I know I can't). 
Try this... https://www.audiocheck.net/blindtests_level.php?lvl=1

That's why I think any audible difference (if there is any) must be coming from extra harmonic content. 
Once sampling rate gets too low or audio frequency gets too high, sine wave input gives following "knobbly sine wave" output from BL3207 ... (sorry for bad quality pics). 

https://1drv.ms/u/s!AvrH61utWEtEi3TYLtqa1xsT9sxz?e=8xT6Rv

while the SAD1024 with 1/2 the sample rate outputs something that does not really look much like a sine wave ...

https://1drv.ms/u/s!AvrH61utWEtEi3O-leEl9cKYEeic?e=eL6BDm

Boxier shape has stronger harmonic content than triangular shape.  In fact I am not even sure it's fair to talk about "BBD gain" of a signal once the output waveform looks different to the input waveform.

Running SAD1024 in parallel mux will no doubt make the SAD1024 output look like the BL3207.

[rankot]

Boxier shape has stronger harmonic content than triangular shape.  In fact I am not even sure it's fair to talk about "BBD gain" of a signal once the output waveform looks different to the input waveform.
Running SAD1024 in parallel mux will no doubt make the SAD1024 output look like the BL3207.

To me, it just seems to be the sampling rate difference. Did you try to run 3207 at 25kHz?

[DrAlx]

Yes, SAD1024 waveform is clearly undersampling the audio.
I have not checked the BL3207 at low rate yet but I'm sure it will do the same thing.

EDIT: I just realised a better comparitive test is to measure signal at the output of my test load (i.e. what corresponds to the "mix point" in the EM) rather than at the BBD output.  I'd probably see waveforms that look closer to sine waves if I did that.

[rankot]

However, could you try to test 3207 running at 25kHz?

[DrAlx]

Tested it.  BL3207 does same thing at 25 kHz as SAD1024 at 50kHz.

Measuring at "mix point" makes the messy waveforms look much more like sine waves (step artefacts are gone) but because the signal is also weaker there it is harder to take good measurements with my simple scope. As far as I can tell, pk-pk voltages from the two BBDs are in just as good agreement as when they were measured at the BBD output (as they should be).

Since the steps in the waveform don't reach the circuit output (the 470 + 47n at the final mix point kills them) it means what I was thinking about harmonics making a difference could not possibly be correct.
Therefore based on my measurements, I can't see how using an SAD1024 vs a BL3207 in this circuit could make a significant difference to the sound, unless there is something I cannot measure (e.g. the noise levels), or if I measured things badly.

Feedback (the color pot) could have an effect since any differences in gain between chips would be amplified.  e.g. the same color pot setting might cause one chip to oscillate but not the other.

In any case, if you want the EM3207 to be a more authentic sounding clone of the original, make the changes I mentioned above and remove Q1.   Without my changes, the wet path will be stronger than an original EM for all frequencies, getting up to over 3.5 dB stronger by 10 kHz as in this picture...

http://photobucket.com/albums/ag185/DrAlx/Graph1_zpsvzhjbxmg.png

That increased BBD gain for high audio frequencies (which is not in the original EMs) has the effect of making the hiss level on the pedal louder too.

[DrAlx]

Also forgot to mention two other resistor changes when you drop Q1.
Change R16 and R18 in Thomeeque's schematic to 13k and 8k2 so that it matches the V6 that the EM3207 is aiming to copy...

http://www.metzgerralf.de/elekt/stomp/mistress/images/1981-electric-mistress-v6-schematic.gif

So the full set of changes is:

• Change (R12, R13, R14, R16, R18, C7) from (4k7, 4k7, 47k, 10k, 10k, 680nF) to (470, 470, 10k, 13k, 8k2, 3n3).
• Remove Q1 and R15 completely.
• Connect the top of C7 directly into C8.

   
   

[Stasss]

I'm planning to build this one.
C8=47nFor82nF?

[DrAlx]

Leave it at 82n so it matches the audio path of the V6 Electric Mistress in the link.

All my tests used a test load based on the V2 EM circuit (which has some different R and C values) but that is not significant.
I would get the same good agreement between BBDs if I had used a test load based on the V6 circuit.
I only used a V2 test load because I wanted to make sure my new measurements agreed with measurements I carried out a couple of years ago using the V2 audio path.

[Ben N]

Is there a "current" (get it?) source of boards?

[DrAlx]

I have not tested a 3007 chip. The data sheet for the 3007 chip shows 0dB insertion gain compared to the 2dB insertion gain of the 3207. I take that gain to refer to the case when load resistance is very high (e.g. over 47k).

Part of the reason I got good comparison between 3207 and SAD1024 was that 10k load on the 3207 chip lowered the BBD gain so that it matched the SAD1024 with 10k load.  That was a lucky find that I did not expect.  I suspect that 10k load on a 3007 chip would give BBD gain that is 2dB lower than the SAD1024. If that's the case then the circuit mods I describe would make a 3007 based EM clone worse (i.e. wet path too weak by 2dB).

BTW, I did not mention this but when comparing 3207 vs SAD1024 with large input voltage (so I could set bias levels) I found that they had similar clipped wave forms at both ends of the clock range.  In other words max input signal level before distortion was similar between chips in this circuit config.

[Stasss]

Thanks a lot DrAlx!
This is very close to the original! Great sounding flanger. I have only one thing to say: It would be great to add a ¨depth¨ control  to have more control. If there is any idea please publish it here.