An Idea

Where we discuss new analog design ideas for Pro Audio and modern spins on vintage ones.
gettestudios
Posts: 23
Joined: Sun Dec 28, 2008 8:47 am

Re: An Idea

Post by gettestudios »

It has been a while.... But i have started down this path yet again. Over the last three years I have meet many challenges to say the least. Although a soldering iron has yet to touch anything on this idea. I am now in a position to explore this further and actually start the build process. I will post the progress here.
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: An Idea

Post by mediatechnology »

Thanks for posting and keep us updated.
User avatar
JR.
Posts: 3700
Joined: Sat May 24, 2008 7:21 pm
Contact:

Re: An Idea

Post by JR. »

In the few years that have passed, this idea appears even less appealing to me, but that is fine.

My project involving DPOTs was put on the back burner by the company I was helping, so I foolishly build my own several channel prototype to answer some questions and prove my design ideas.
am_img.gif
am_img.gif (48.21 KiB) Viewed 12416 times
Obviously there is a lot more on the proto than just the gain stages. The DPOTs are the tiny area on top right corner, VCAs across the top left.
left edge is parallel input and output Talleys, and programming pins.
Bottom edge control pots, and dome switches, LEDs to mimic several channels of AM.
Right side was audio I/O

Since I was at a crossroads regarding VCA vs DPOT as variable gain elements for cost vs performance, I did several approaches in parallel on the proto. Since the project appears to be on indefinite hold, I guess I can share some of my design thoughts. The product was a digitally controlled analog Automatic Mixer, a class of automated mixer that shares the gain between multiple mic channels so that the total gain of all channels combined is the same as just one open mic. This reduces the sensitivity to feedback for multiple talker applications.

My design effort was both to be cost effective, and superior sonic performance. While the new THAT VCAs are quite good, they are not perfect so there is always room for improvement. The DPOTs in theory are just switched resistors so in principle could be quite low distortion, while the DPOTs generally suffer from reduced voltage swing (not so much in expensive variants) and limited step size (8-12 bit resolution).

A quirk of automatic mixers is that most of the active important gain changing is up around unity gain, since channels get turned down by other channels being turned up, so signal quality and step size at lower gains are not very apparent because of the masking by louder channels.

One strategy to improve VCA performance was to make it subtractive. Since I only needed unity gain or attenuation, I could engineer a topology where the VCA was completely out of the circuit at unity gain, with signal supplied by a parallel resistor, then the VCA gets turned on to provide attenuation. This way the noise and distortion added by the VCA was only present during significant attenuation, and since the AM is attenuating signals because they are smaller, the distortion then is modest.

My strategy to deal with DPOT limitations was to pad signal 6 dB before the DPOTs and synchronize my gain updates to zero crossings. SInce I had to capture and measure signal level for the automatic mixing , detecting zero crossings was just another microprocessor task. Another strategy to reduce apparent zipper noise was to use HF pre/de-emphasis before and after the DPOT so the clicks get LPF by the de-emphasis. In general design this costs HF headroom, since the DPOTs need some padding already I could roll the pre/de into the pad make up gain and only lose the HF headroom, generally not a problem with real world signals and modest emphasis.

I got the platform running and tried the DPOTs first, because they were simpler and cheaper from a glue circuitry perspective (just push digital gain info to the DPOTs). The VCAs needed DACs so the micro could send control signals and convert to voltage with some filtering glue for that. Opamps and precision resistors for the subtractive circuitry, etc.

A lot of my coding was to make the AM gain sharing algorithms operate over decent dynamic range from cheap A/Ds (12 bit) built into the micro. I fired up the proto with DPOTs first, with no zero crossing sync working, just changing the gain at a fixed sample/update rate, following the signal changes. I used the audio from my TV monitor for one mostly speech input, and a couple channels from my MAC/PC to demonstrate automatic gain changes with level.

I was pleasantly surprised buy how good the DPOTs worked with no zero crossing and no pre/de emphasis. So never bothered to fire up the VCAs while I know another engineer who used the subtractive VCA approach to make a high quality limiter.

I stopped working on this when my friends advised me that they back burnered this product. I remain even more convinced now that I could do this cheaper and better, fully in the digital domain, but that was inconsistent with their company's market identity as premium analog gear. I have a product guts mostly designed gathering cobwebs, but that is OK, I got to stretch mentally along the way.

====
Admittedly some of my design above is more optimized for the very specific AM application, so take that with a grain of sale.

based on what I know,to make a D controlled A console I would be inclined to use DPOTs for all variable eq filters. and mostly set them and forget them gain stages like aux sends, gain trims, etc. VCA's may still justify their keep for channel faders, using my current source summing for decent bus performance. With a micro brain available there may be other tricks available to scale VCA operating points for lower distortion, etc.

I still don't see this as a very commercial product. In the meanwhile Behringer has popped their <$3k digital console that apparently doesn't suck (from what I read). So the digital evolution continues.

Of course the customer is always right, and it's only (your) money and (your) time, so enjoy..

No< I'm not inclined to publish schematics of this because the product could still get reincarnated and i still think it is viable, while just barely to customers with analog fixations. The prejudice against digital seems to be fading more and more.

JR

PS: For mental masturbation I have also looked at an all DPOT sum bus topology to make pan and level mix, using the DPOTs to make synthesized current sources for very low bus noise gain. This approach uses something like 4 DPOTs per stem (or maybe more) so is not very cost effective but might deliver Dangerous performance levels but with digital control of level and pan, hopefully to interface with some generic studio automation software.

PPS: I haven't looked at tweaking out DPOTs for uber-performance on the bench, but I suspect buffering the DPOT wiper could reduce any non-linearity from the wiper switches modulated by load current. I don't have the time to pursue uber-circuits and don't have a bench good enough to parse out extreme performance.
Cancel the "cancel culture", do not support mob hatred.
User avatar
mediatechnology
Posts: 5444
Joined: Sat Aug 11, 2007 2:34 pm
Location: Oak Cliff, Texas
Contact:

Re: An Idea

Post by mediatechnology »

Very cool John, thanks for sharing.

What step size did you try with the DPOTs?
I realize that the DPOTs were probably "linear taper" but around 0 dB gain what was the approximate step?
juniorhifikit
Posts: 139
Joined: Sun Oct 04, 2009 8:37 am
Location: Oakland/Paris

Re: An Idea

Post by juniorhifikit »

Very curious about the circuitry to limit the Dpot moves to the audio's zero crossings...
gettestudios
Posts: 23
Joined: Sun Dec 28, 2008 8:47 am

Re: An Idea

Post by gettestudios »

JR- Thank you for a very detailed response. i understand the belief that Digital is at this point taking completely over, However as long as we continue to record Music or a linear signal, analogue will always be a part of the signal chain. As of right now, microphones still use an analogues way to capture an acoustical sound wave (movement resulting in voltage). Voltage is a physical medium that can still be felt by us mere humans (if the voltage is great enough...). Were as Digital although a physical representation in 0's and 1's (voltage hi/low) changes the wave form radically when compared to the original source. Although the change really starts becoming minimal at higher sampling and bit depths. So the question is: Is it worth the time and resources to venture down a path that many believe is quickly becoming obsolete? Certainly I will have varied responses to that question depending on were each person stands on the technology involved.

So, lets step aside from preference for a moment and look at the current state of the market today versus a few years ago. PT is by far still the front runner, followed by Cubase/Nuendo variants and Logic Pro. Clearly, the DAW has completely dominated the market space. However, the Big boys are still manufacturing and selling analogue consoles. A recent trend rising, is the use of analogue summing mixers (API, Shadow Hills, Etc.) to sum multiple outs of a DAW to get the feel of an analogue mix bus. So the statement analogue is fading and counting its days out and Digital is the new standard, simply is not a correct statement. The better question to ask is why, why is digital were it is and why analogue has not faded away like the cassette tape or 8 track cartridges. The answer is simple. Analogue still holds something that our human ears refuse to let go (Put magic word here) However, what analogue lacks digital fulfills in spades, its Information. That simple, the amount of information provided to the end user on each signal is massive. When compared to an analogue desk (meters only) you begin to understand the bigger picture. As engineers we thrive on versatility and information, the more we know about said signal and how we are altering it, the more we feel in control of the mix. If we were to take an honest look at mixing totally in analogue and compare the feedback of information and versatility to that of digital, one could compare it to driving blind down a predetermined path.

There you have it, the premise and goal of my project. An analogue console with the informational feedback and versatility of a Digital console. Peak anyones interest? :D

Joel
Last edited by gettestudios on Sun Mar 10, 2013 3:10 pm, edited 1 time in total.
gettestudios
Posts: 23
Joined: Sun Dec 28, 2008 8:47 am

Re: An Idea

Post by gettestudios »

juniorhifikit- Here is a link on a paper by AD on Digipots that include a Zero Crossing circuit (page 42)

http://www.wdv.com/Electronics/PaperCir ... final1.pdf

Hope that starts to shed some light on the topic for you.


Joel
User avatar
JR.
Posts: 3700
Joined: Sat May 24, 2008 7:21 pm
Contact:

Re: An Idea

Post by JR. »

mediatechnology wrote:Very cool John, thanks for sharing.

What step size did you try with the DPOTs?
I realize that the DPOTs were probably "linear taper" but around 0 dB gain what was the approximate step?
As I mentioned most of the fine gain detail automatic mixers is needed up around unity gain or full scale, which on a linear DPOT is where 1 LSB change is too small for me to calculate in my head.

Only near the bottom of the gain range does the gain steps size get grainy, but who cares about step size when you are turning down a signal that is already quieter than the active channels. Sp step size to a signal at -30, when you're tunring it down another 30 dB is not audibly significant.

Perhaps unacceptable for master fader fades to silence, but in a complex mix, not so much.
-----
It's been a while but the DPOTs are linear and I performed my gain algorithm calculation such that unity gain was digital full scale, and proportionately less gain was the digital fraction of that digital word, so output fell out of my calculations already scaled directly for a simple serial push to the DPOTs.

Performing the Automatic mixer using VCAs was far more complicated, since VCA want to be driven with log voltages, I came up with an algorithm to convert my linear digital gain to log, requiring I loop through per bit of output resolution. This was bits on the DAC used to drive the VCA, likewise the bits scaled down by factor of two per bit. So if MSB was 24 dB, next 12 dB, 6,3,1.5, 0.75 , 0.37, I am too lazy to look up what I used but 10 bit DAC delivers small fraction of a dB resolution. I also pushed the microprocessor up to 20 MHz clock so even if my log conversion took 20-30 clock cycles that's all of a microsecond and a half. Further I could selectively skip calculating the lower bits of resolution for signals receiving significant attenuation, because again, the resolution there isn't as apparent.

Since the DPOTs worked so well, I never got to find out if there were extra snakes in the wood pile with driving the VCAs that I didn't anticipate. Note: smoothing on the DAC output means VCAs would not experience even the small sharp gain steps.

JR
Cancel the "cancel culture", do not support mob hatred.
User avatar
JR.
Posts: 3700
Joined: Sat May 24, 2008 7:21 pm
Contact:

Re: An Idea

Post by JR. »

juniorhifikit wrote:Very curious about the circuitry to limit the Dpot moves to the audio's zero crossings...
No extra circuitry was involved. I was already sampling the channel audio to do my gain calculation, so I just needed to keep track of when the sample data changed from negative to positive, or vice versa.

I came up with a pretty involved SPI service routine. I rolled my own SPI code with multiple different clocks and data pins, and latch lines, so I could push, several sets of serial data at the same time by multiplexing.

i.e. set up new data on multiple data lines, than switch clock one. This moves that data into all chips using that same clock. Then reset the data lines for next bits for next group and switch the next clock, etc. When you have pushed 16B or whatever data, you set the latch line to latch it in. You can actually push false data through a device,and as long as you don't latch it it harmlessly falls out the other end. I have already forgot several small details

I used 4x16 line LED latches and they can be daisy chained output of one and into input of the next so i could update all of them with 64 spi clock ticks. I didn't need to light 64 LEDS (they were handling the meters separately) but I used the LED latche lines for my output tallies, and even to strobe channel switches for input switch detect multiplexing. The DPOTs were stereo or duals. The Automatic Mixer had 6 mono inputs and 2 stereo inputs (so 5 duals, but only 8 sets of unique gain data). I pushed the same gain to L and R sides of the stereo channels. The dual DPOTs used one bit in the control data to select A or B, so I organized the Stereo DPOTs so one stereo channel used both A sections, and the other channel both Bs, so I could run them in parallel and both L/R paths updated gain simultaneously.

While I never got to use it, my general plan for zero cross synchronization, was to use a smart algorithm to weigh between largest gain update change pending, and next anticipated zero crossing (based on prior average between ZC and how long since the last ZC). Since I was pushing gain updates to multiple DPOTs at the same time only the loudest/largest gain change was getting synchronized. I could either push old gain to other channels or tolerate the zero cross hit if gain/level was low. Several moving parts to keep track up but manageable with enough microprocessor horsepower.

The prototype sounded so good I never had to implement the Zero cross synchronization, but basically I would push the data to the DPOT and wait for the A/D to detect the =zero crossing and then set the latch to update the new gain.

Doing this zero cross sync externally with fast comparators and glue logic is a PIA, while not heavy lifting to perform inside a 20MHz micro, and like I mentioned you can use non-linear decision making to make it smarter/better. If needed.

This is another reason why i ran out of patience waiting for them to build a prototype (actually in their court to do). I had too many unresolved questions about the technology, so I just said screw it and made one myself.

+++++++

I only started messing with DPOTs a few years ago, my judgment is that the modern DPOTs are a lot better than early generation DPOTs (they've been around for decades). My research turned up old reported issues with major control bit transitions Like changing from 0x7F to ox80 was far more than a 1 bit step, with charge perturbations due to internal architecture. Zero crossing synchronization cures the major bit transition issues, in addition to the obvious audio step change.

Ironically perhaps doing this all in the digital domain is actually easier.

JR
Cancel the "cancel culture", do not support mob hatred.
User avatar
JR.
Posts: 3700
Joined: Sat May 24, 2008 7:21 pm
Contact:

Re: An Idea

Post by JR. »

gettestudios wrote:juniorhifikit- Here is a link on a paper by AD on Digipots that include a Zero Crossing circuit (page 42)

http://www.wdv.com/Electronics/PaperCir ... final1.pdf

Hope that starts to shed some light on the topic for you.


Joel
If you are talking about the circuit on page 42, there are several things that I don't care for, or don't completely follow their reasoning.

The opamp buffer U6 is sitting at about, +5 mV DC. This dumps right into the DPOT through R4 and R5 so for the values shown the DPOT has 2.5mV of DC across it all the time at idle.

The comparators create a window from roughly +2.5mV to +100mV when both comparators are high, so high at output of AND gate.

The DPOT accepts data when the chips select line is low, then latches that data when this chip select line goes high again. So the controller pushes new gain information with CS low, when finished the CS line goes high then waits for both comparators to go high. When everybody is high and happy the CS line on the DPOT goes high and latches the new gain data.

For audio coming from negative polarity towards zero this will update at 2.5 mV or pretty close to zero, but coming from the the positive direction it latches when it drops below 100 mV. Not very close to zero. If they need 100mV window to reliably latch at high audio frequency, I'd prefer making the window symmetrical around 0V.

At least they get points for using all Analog Devices parts. :lol: They got the concept right and maybe with different circuit values the window was more symmetrical around 0V. I still don't like the DC across the DPOT.

JR
Cancel the "cancel culture", do not support mob hatred.
Post Reply