MUFF WIGGLER Forum Index
 FAQ & Terms of UseFAQ & Terms Of Use   Wiggler RadioMW Radio   Muff Wiggler TwitterTwitter   Support the site @ PatreonPatreon 
 SearchSearch   RegisterSign up   Log inLog in 
WIGGLING 'LITE' IN GUEST MODE

modding the Plan B Model 9
MUFF WIGGLER Forum Index -> Music Tech DIY Goto page 1, 2  Next [all]
Author modding the Plan B Model 9
intellijel
The Plan B Model 9 mixer is a great concept and layout but it has the major issue of channel bleed. The actual design is very simple (and would take no time to reverse engineer, it is textbook mixing circuits).

Has anyone looked at modifying one of these to eliminate the channel bleed issue? I think that this may not be that difficult for someone more savvy than me. (I have tried the Peter G suggested Capacitor mod but it only made a minor improvement).

If this was fixed it would definitley become an essential module to own.
wetterberg
Judging from what I've seen and heard this last year or so I reckon it's a simple mistake that *can* be rectified.

Some pics? Maybe the tech-savvy among us can see what's going on there.
emergencyofstate


jonkull
Focus?
emergencyofstate
sorry, my iphone sucks at up close focus (obviously) I'll try to get one of the photogs at work to take a pic...

Edit* replaced with way better photos. have a look above.
chinard
ok i know that there is a mod listed on peter's yahoo group where you swap the 4 blue bypass caps with a higher value.
It is an improvement but it does not completely eradicate the problem and has apparently been applied to all recent model 9's

http://launch.groups.yahoo.com/group/PLAN_B_analog_blog/message/2269

Quote:

Model 9 Crosstalk Mmprovement

Hello to everyone.

This is an easy fix. On each of the two ICs on the M9 (one a TLO74, One a TLO72), you will find two small blue capacitors flanking those components which are unmarked (meaning there is no C number - as in C1, C2, etc. - designator).

SO two per IC for a total of 4 caps.

On the M9 all four of those guys are by the edges of the PCB. There are a few other small blue caps on the board, but those have designator numbers and are in the pf range, somewhere around 22pf (I'm at my g/f's place, not around my schematic reference or system, please pardon my 52 year old memory). These shouldn't be changed. They merely kill potential oscillation in the feedback loops of the gain buffers (wouldn't ya know it, hardest thing in analog is to get a oscillator to do it's thing properly, yet it will happen all over the place were you don't want it so you have to through components to it to stop that from happening).

Until quite recently those four caps were .01 uf, marked "103" on the cap itself. They need to be removed and replaced with .1 mf, which is marked "104".

That's it. Not that bad, five minute job and will cut the crosstalk down a lot. We've been shipping the m9 with this new part since this was discovered about a month or so ago and are now using .1's on all of our modules.

For a reference of how to read the value of small capacitors, see this:

http://www.electronics2000.co.uk/data/itemsaf/capconv.php

CAPS are marked with the Picofarad value. A .01 microfarad is 10,000pf - their marking reads 103, which visually must be read as 10 plus 3, meaning three extra zero's after the initial 10 (10,000). .01's = 100,000pf
or 10 plus 4 - four extra zeros after the 10.


I think eveyone here agrees that this module would be spectacular if the crosstalk were completely eradicated.
consumed
chinard wrote:
I think eveyone here agrees that this module would be spectacular if the crosstalk were completely eradicated.


that, and nice switches, would make it fully fantastic.

to the original point, to really figure out how to make the design better would require schematics. since we know those arent available, someone would have to work their way through the circuit and create the schematic by hand, in order to look for areas of improvement, which is not a particularly simple task. its also possible that something in the design itself (where the switching occurs in the circuit, or something else) might be the source of the symptom, which would probably not be as simple as a capacitor switch out. this is where someone with experience in designing and testing circuits would really be able to help. even then, the amount of work necessary to make it crosstalk-free might be more trouble than its worth...

the crosstalk on my model 9 is pretty low. for audio use, its mostly acceptable. for CV use, i would probably perceive the crosstalk easier. but i dont run particularly sensitive CVs through it anyway (like CVs going to 1v/O inputs). the idea of having some crosstalk in a mixer bugs me a little, or when im testing in an isolated environment i can notice it more, but when a patch is in full swing i cant tell. so im not bothering with the cap mod for now.
intellijel
I did the mods suggested and their was a minimal improvement. If you are mixing audio after the VCA you can clearly hear all other sources (some that aren't even patched in) even if they are all muted.

The circuit is right out of first year EE and opamps. It would take about 10 minutes to trace it out on a napkin.

Maybe that should be my next step....
consumed
intellijel wrote:
The circuit is right out of first year EE and opamps. It would take about 10 minutes to trace it out on a napkin.

Maybe that should be my next step....


you should go for it!
intellijel
btw, this is basically the theory behind the design with switches added to alter routing (you will notice some similarities to the m9 photos):

intellijel
p.s. that image came from here:
http://www.all-electric.com/schematic/simp_mix.htm

Just reading about using virtual earth ground to eliminate channel bleed...
Tim Stinchcombe
intellijel wrote:
Has anyone looked at modifying one of these to eliminate the channel bleed issue?
Yes, I have reverse-engineered it to see what is going on, and how the bleedthrough might be improved. The short answer is that some improvement is possible - however if my asessment of the problem is correct, to properly eradicate the issue would require completely re-laying out the PCB.

The circuit is relatively straightforward, 6 op amps in total: a summer followed by an inverter for both channels A and B; both channels are then added and the sum also inverted.

The first thing I noticed was that the channel A summing op amp was oscillating like crazy, at around 280kHz. Strangely, unlike the inverter which follows, there is no stability cap in the feedback path. I tried several other TL074 chips, a TL064 and a TL084 - all of these also oscillated, but to a more or lesser degree, depending on type, and made me think it was a general problem, and not just my module. One I did try which didn't oscillate was an LM348 (a quad 741 clone). I haven't yet investigated if adding a cap quashes the oscillations, as I stumbled across something that did: the summer has a resistor to ground on the non-inverting input - ostensibly to counter the offset due to input bias current at the other input - but in any case the value is badly chosen, and the input bias current is so low to me it seems completely unnecessary. Hence I shorted it out, and to my surprise the oscillations vanished. From my (as yet incomplete) understanding of the situation, this tends to point to lots of stray capacitance around the circuit.

[Edit 4 years later!: I do now understand how the oscillation arises, and have completely revamped the page on my website to include a good deal more detail:
http://www.timstinchcombe.co.uk/index.php?pge=m9]


With this out of the way it was easy enough to see the cause of the bleedthrough - there is enough stray capacitance around the board for the signal to capacitively couple to the channel not being switched through to, which then acts as a differentiator - basically a highpass filter, which lets more and more of the signal through as the frequency is increased. Given the way the tracks run up-and-down the board to all the switches, and the ribbon cable connecting the boards (though with every other wire grounded, that shouldn't be an issue...), this doesn't seem too far fetched, and from simulation and calculations I estimate it to only be about 5pF, but that is apparently enough to cause the problems.

So, one means to alleviate the problem would be to reduce the value of all the 100k resistors used in the summer op amps: if they were all made 10k say, then the summing would still be unity gain, but the differentiator affect would have its gain dropped by 10 (i.e. 20dB) and so that would help. This would required replacing 7 resistors on the board. I could probably whip up an annotated photo if anyone fancies giving it a go, and I'll also check back through my notes to see if there is anything else I haven't mentioned that might be worth doing so.

Tim
intellijel
nice work!!

I will definitely give this a go when I get home tonight.

The 7 resistors are probably one group of 3 and one group of 4 right? (I think I can see them from the photos above).
Tim Stinchcombe
intellijel wrote:
The 7 resistors are probably one group of 3 and one group of 4 right?
Not quite! I remembered I had actually scanned the boards, as I spent some time actually checking how the traces themselves criss-crossed each other. Hence it was a simple matter to annotate it, see attached. I've ringed in blue the 100k's to replace with 10k's - 6 on the main board, the seventh on the jack board (which will be tricky - probably easiest is to cut the old one off, and use the 'stub' wires to solder new one onto, making sure the end of the jack plug doesn't interfere with it!). The three I'd recommend replacing with a direct short are ringed in red.

Tim

P.S. Batz's 'simple virtual earth active mixer' circuit is of course closest to the M9.
intellijel
I did the mod....... WORKS!!!!

Almost ZERO bleed!!!! I had it cranked to the max, listened on headphones with two saw waves (one from an AFG and one from a M15) plugged into seperate outputs. I did notice that if I put the hottest possible signal out of a filter (z2040) that it did start to bleed a tiny bit but backing off the filter gain a little bit eliminated that problem.


This just became the best mixer you can buy (after 15 minutes worth of mods)
paults
If the photos indicate that is a 100K resistor from the + input to ground, that is indeed way too high. Replace it with 1K (instead of a short) and that should help the DC drift but also not oscillate.

Also: the sockets used are not helping matters any. They add series inductance and with at 100K hanging off the high-impedance + input you could probably pick up AM radio It's peanut butter jelly time!
Tim Stinchcombe
intellijel wrote:
I did the mod....... WORKS!!!!

Almost ZERO bleed!!!!
I'm pleased - that extra 20dB attenuation is worth having! However you probably can still hear some bleedthrough at higher frequencies. The problem seems worse on Channel 4, and also when the input pot is fully clockwise (and of course with the switch in the central off position) - at higher frequencies, >10kHz or so, I expect you will still be able to hear the input signal at the output, even though it is not switched through.

I dug out a couple of traces I took early on, and reviewing them re-convinced me that the capacitive coupling is the problem: the (blue) scope trace is with an approx 1.6kHz square wave input, and we see a typical differentiated waveform out (this was with an LM348, which didn't oscillate); the (red) simulation trace is with a 4.7pF cap only at the input - I think the simularity is pretty convincing.

The last frequency response plots show the simple highpass differentiator effect with a 4.7pF cap, and with the feedback resistor at both 100k and 10k. This shows the 6dB/octave decrease in attenuation as the frequency increases, i.e. that we get more bleedthrough as f goes up. This ties in quite well with measurements I took: at approx 10kHz, -32dB for 100k, and -51dB for 10k; at approx 660Hz, -54dB with 100k (and I couldn't get a sensible reading using 10k's here!).

Tim
intellijel
so is there anything else that can be done to eliminate the capacitive coupling problems? Or is this as good as it can get?

Here are some suggestions I got from the SynthDIY list:

Quote:
I had a look on the plan B website and the way they do the routing leaves the signal disconnected, not shunted to ground, when in the mute position. Maybe the switch is bleeding the audio through. You could try switching it to output B and see if there is any audio on output A...

Perhaps a lower impedance path around the routing switch would help (if that is the cause of the problem).


Another solution would be to use a double pole switch for the channel muting and use the other pole to control FET's or analog switches to shunt the signal to ground when in the mute position.
Plan B
Hi guys.

I really appreciate the time you've taken to suss this out. Nice work, I'm serious. Remember - lowering the summing resistors is going to raise the current draw proportionate to Ohms law. 1k's I would avoid.

The likelihood is the switches aren't the problem.

The removal of the 100K baising resistors may assist in quelling some of the bleed, but you's now introduced a variable which will case less than accurate outputs for DC applications - when using the the 9 for mixing control voltages. If this isn't a problem for your specific application, cool.

One thing you may want to try is replacing the TL074 quad op amp with something pin compatible in the Analog Devices OP series, like the OP11 and 400 - which is going to have better isolation form amp to amp which I believe is the real demon here. Down the line i may redesign the 9 with two 72's instead of a single quad amp as the dual I've found doesn't have this tendency.

Regarding op amp oscillations: first thing always in checking this is making sure you've got a god path to gr0ound on your scope probe - using the clip on the probe end to a nearly ground node on the board form the location you're measuring. If this guideline s followed and you're seeing this, raise the value of C1 and C2 - both should be 22pf. If you're using the 9 primarily as an audio processor, you won't hear a huge difference with higher values. With DC you're going to run into issues, as you've just introduces an integrator into the circuit - a portemento basically.

hope this helps, feel free to contact me offline at peter at ear dash group dot net.

- P
Plan B
regardng my last post:

A god path to ground may do the trick, but a 'good' one will suffice if the Allmighty is off to bigger tasks at the time.


...sorry for the typos.

- P
intellijel
Plan B wrote:
Remember - lowering the summing resistors is going to raise the current draw proportionate to Ohms law. 1k's I would avoid.

I used 10K so the current draw will be 10 times more per source but that should not be a big deal.[/quote]

Plan B wrote:

The removal of the 100K baising resistors may assist in quelling some of the bleed, but you's now introduced a variable which will case less than accurate outputs for DC applications - when using the the 9 for mixing control voltages. If this isn't a problem for your specific application, cool.

So what about Paul's suggestion to put a 1k there instead?
Tim Stinchcombe
intellijel wrote:
so is there anything else that can be done to eliminate the capacitive coupling problems? Or is this as good as it can get?

Here are some suggestions I got from the SynthDIY list:

Quote:
I had a look on the plan B website and the way they do the routing leaves the signal disconnected, not shunted to ground, when in the mute position. Maybe the switch is bleeding the audio through. You could try switching it to output B and see if there is any audio on output A...

Perhaps a lower impedance path around the routing switch would help (if that is the cause of the problem).


Another solution would be to use a double pole switch for the channel muting and use the other pole to control FET's or analog switches to shunt the signal to ground when in the mute position.
Yeah I saw that. When I was looking into it, I did try grounding the signal at the switch, and it does help, giving another couple of dB or so. Here is the dB 'budget' from my notes (most at 660Hz, I think, and with such small measurements, handfuls of millivolts, sometimes the readings get a bit 'subjective'...):

As it is (minus the oscillation!), input to output: -53dB

Change in:
a. decouplers 10nF -> 100nF: -2.5dB
b. grounding switch common/middle tag: -2.8dB
c. removing ribbon: -1.8dB
d. 100k -> 10k (@10kHz): -19dB

Switch middle tag desoldered, input direct to it (i.e. bypassing pot+resistor): -67dB - this is an overall 'in to out' figure, and I think shows lots of the coupling occurs upstream of the switch.

My conclusion kind of was, yes, you could work quite hard at doing various things, each of which will give another couple of dB, but in the end your effort would probably be better directed to laying out the PCB again from scratch, avoiding the problems in the first place... (and maybe duplicating the input resistors and placing the switches the other side, further away from the virtual ground? I dunno, I'm sure there are others better at PCB layout than I am, who could do the job properly!)

Tim
Tim Stinchcombe
Hi Peter,
Good that you've joined us, but be warned, I can have a fastidious approach to relevance and accuracy!

Plan B wrote:
1k's I would avoid.
Yep, that's why I suggested the 10k's, which I think we can get away with.

Quote:
The likelihood is the switches aren't the problem.
Agreed, this is sort of implicit in the -67dB figure of my last post (which crossed with yours).

Quote:
The removal of the 100K baising resistors may assist in quelling some of the bleed, but you's now introduced a variable which will case less than accurate outputs for DC applications - when using the the 9 for mixing control voltages.
With the typical input bias current of the TL074 (at 25C) given as 65pA, the voltage across a 100k is a mere 6.5microvolts, so I'd be astounded if anybody could notice the loss of these resistors - if you could, you'd be much more concerned with the (in comparison) whopping great 3mV of input offset voltage (and hence probably wouldn't have chosen TL074s in the first place).

Quote:
One thing you may want to try is replacing the TL074 quad op amp with something pin compatible in the Analog Devices OP series, like the OP11 and 400 - which is going to have better isolation form amp to amp which I believe is the real demon here.
I'll try and get something like one of those, and try it, but on the evidence so far, I'm betting it is stray capacitance due to board layout which is the main culprit.

Quote:
Regarding op amp oscillations: first thing always in checking this
is making sure you've got a god path to gr0ound on your scope probe - using the clip on the probe end to a nearly ground node on the board form the location you're measuring. If this guideline s followed and you're seeing this, raise the value of C1 and C2 - both should be 22pf.
This isn't an issue here - the oscillation is clearly visible 3 op amps away from the one oscillating! The problem is lack of a stability of that op amp.

Quote:
If you're using the 9 primarily as an audio processor, you won't hear a huge difference with higher values. With DC you're going to run into issues, as you've just introduces an integrator into the circuit - a portemento basically.
This is moving into an area in which I'm less well versed (i.e. the impact on musicality), but again the arithmetic leads me to question whether there is likely to be an issue here. The 47pF (as fitted at C1 & C2 in my M9) with the 100k indeed gives a lowpass filter, but whose cut-off is around 34kHz - simulation suggests a 5V step change input would have a rise time of about 30us; even if the cap were ten times bigger, this would still only be perhaps 300us. If driving a VCO, would the ear really be able to distinguish the difference between the speed of change of pitch at such rates, or would both simply sound like an instantaneous change in pitch? (And in any case going from 100k -> 10k increases the cut-off 10 fold, countering any effect of the stability cap.)

Tim
paults
Without a schematic, I'm just guessing here. But there may be another issue: long pc board traces from the 4 summing resistors back to the op amp (-) pin.



From this angle, these 4 resistors appear to go to the wipers of the pots. I will then assume this is the summing node for the op amp (1/4th of the TL074).

This trace/layout will be very prone to crosstalk due to the long length and capacitive coupling pointed out by Tim. The only way to reduce this is by dropping the input resistor values. I would start with maybe something on the order of 4.7K.

I don't know what the TL072 is "for", but it appears the feedback resistor between pins 1 & 2 is quite a distance away (because the connector is crowding the part). Again, pin 2 of this part is probably a summing node and this node will be prone to noise pickup/crosstalk.

Lastly: I will bet that putting in a "better part" (usually meaning faster slew rate and larger GB product) would make matters worse. Tim's suggestion of a LM348 (essentially a quad LM741) is an old 'trick'. In fact, it's used on the DKI Synergy output DAC filter for this very reason. I stuck a really good ADI quad part in there and the noise floor tripled.
daverj
Quote:
The removal of the 100K baising resistors may assist in quelling some of the bleed, but you's now introduced a variable which will case less than accurate outputs for DC applications - when using the the 9 for mixing control voltages. If this isn't a problem for your specific application, cool.


I don't have one of these modules, so can not look at the design, but based on what I've been reading here I would say the above quote is not accurate.

If the circuit is using TL07x op amps, and the circuit is an inverting amplifier with these "bias resistors" on the non-inverting input, then their value is not only incorrect, but they are not needed at all.

Op amps using bipolar input transistors, such as the 741, require matching the input impedances to compensate for input bias current. In a normal inverting amplifier circuit the bias resistor on the positive input is equal to the combined impedance of the resistors on the negative input. With a 100k feedback and one 100k input resistor, that would mean a 50k bias resistor on the positive input. With 2 input resistors, the bias resistor would be 33k, and with 4 it would be 20k. (assuming all resistors on the negative input are 100k)

However, this technique is only used on op amps with bipolar input transistors (old amps, like the 741 and newer low noise audio amps that use bipolar inputs). JFET op amps such as the TL07x series do not need the bias resistor. They have input bias currents on the order of 4,000 times lower than the 741. In fact, having that resistor on a JFET op amp can add instability in many cases, if the layout is poor and the stray capacitance values are high.

That is why the manufacturers of JFET amps like the TL07x suggest connecting the positive input directly to ground in an inverting amp configuration.

But, as I said, I don't have one of these to look at, so don't know the actual schematic of this module.
MUFF WIGGLER Forum Index -> Music Tech DIY Goto page 1, 2  Next [all]
Page 1 of 2
Powered by phpBB © phpBB Group