AD8310 - anyone played around as an audio level detector??

Yeah, I know it was primarily designed as an RF part for a RSSI, but it does seem to give a really fast (peak) level detection that is logarithmic (linear in db) scale.

It's been out a long time, but I just really noticed it and thought seriously about what it might do in audio specialty applications where fast attack might dodge artifacts.

Anyone played around with the chip??

Looks like an RF part... Didn't know speed was an issue with log conversion?

JR

> fast attack might dodge artifacts.

?? Seems like fast-attack would *show* problems you could never hear.

> Looks like an RF part... Didn't know speed was an issue with log conversion?

The classic junction-around-an-opamp converter gets real slow at low input levels. In the early 1970s, a bare-face implementation was likely to be 5KHz at -50dB. (Main problem is the loop must be compensated for high current then the gain falls off at low current, dropping GBW.)

With today's faster opamps it should not be an issue in the 20KHz band.

Not that I ever really needed full bandwidth at lowest level.

Yes, I look at things like AD8310 (there's several, mostly from Barrie Gilbert). Some are DC-coupled so should pass audio. Some early radio-oriented chips could have a few dB error over the range, but some recent ones are as good as any junction-log scheme.

The main issue for broad use: it's a $5 part, versus $1 for a junction log (and now probably a PIC). Since audio companies never rise far above bankruptcy, $4 per channel is no joke.

Thanks for the feedback, guys.

PRR - You're right on target about the price - I'm sure I can pick less pricey ways to get the really fast attack I'm after (follow the signal right up the initial transient) and get the actual level detection in more conventional ways. Back to pre-detection EQ...

I should have been clearer to start with...The "artifact" I'm trying to dodge isn't the attack time for a mixing console compressor - which I should have stated in the intitial post - it's the sound of hyper-abrupt noise reduction in an audio "swell" effect, "tape-reverse" effect, or whatever you want to call it, for electric guitar. I just want to silently suppress the pick attack. No click, pop, nothing.

It's good to get grounded about the cost - the design is not for a commercial product, but I can tend to get infatuated with what an amazing feat of engineering a part like the AD8310 crams into a MSOP, and get caught up looking for a way to take advantage of it somehow<sheepish grin>

Just wondered if anybody around here had found it useful in some audio app...prolly not, huh? <grin>

Oh well...back to the breadboard!

If I get it right you want some effect like boss slow gear, if so you probably don't have hurry and you don't need any fastest there, just make your initial state muted and then going up as slow as you whant, or as fast as you like. A not so fast Log converter should do the job, don't think in a compressor, think in a gate, where initial state is mutted and then goes up to 0dB gain or whatever. This way you  can't have any artifacts or noises because signal is mutted, you need feed foward, no chance of doing so feed back but if you are looking for a log converter I guess you are going to use a VCA and feed foward is not a problem.

JS

Just to be clear, IMO rectification is the dominant process that suffers from speed issues (gain bandwidth) at low level not the logger.

I became all too aware of this designing the dB conversion for the TS-1 back in the early '80s. The vast majority of rectification circuits log or linear, require the active (opamp) gain stage to swing from - to + a full diode drop at every zero crossing as the signal passes through 0V dc.  Generally opamps that start out with 100dB of open loop voltage gain, this input voltage to make +/- .5V at the output is no big deal. However at higher frequency as this open loop gain gets rolled off for stability compensation gain becomes an issue.

To put numbers to this, Say your general purpose opamp has been compensated down to 60db of open loop gain by 20kHz. This means at 20kHz to make the 1Vp-p output swing at zero crossings, the input needs 1mV of drive. This 1 mV is effectively subtracted from the input signal. For a several volt signal this 1 mV is not missed, but for an input audio signal that starts out with only a few mV, subtracting one mV is significant. In practice this looks like a LPF rolling off the audio but instead of just being frequency dependent, its affected by both level and frequency.

This is not a new or unknown problem so there are sundry circuit techniques to improve this metric. For me it was a matter of concern because I was making test equipment so little things like frequency response errors were undesirable. In my experience for dynamics processors the frequency response of the side chain at -60dB is not a big issue. It's not like it stops working at -60dB but the HF response falls off.  I found when designing companding NR paths for tape recording or effects, it was often desirable to frequency shape the side chain to avoid those pesky extreme frequencies that don't always survive the media (tape, BBD delay, etc).

Getting back on topic, the response of a logging junction will be somewhat affected by current density, so that becomes a tradeoff to get useful dynamic range. Too much current causes errors at the high end due to Rbb resistance, which adds a linear term to the log output voltage. In my TS-1 the rbb linear error was almost 1dB at +20 dBu.

I suspect the RF chip is using an alternate approach. I would need to see more details about the inner circuitry to comment intelligently but I can tell from the block diagram it isn't a simple log conversion.  So yes it promises to be faster, I am still unsure that this is useful for audio processing. 

@ Mr. Coffee methinks your problem is more fundamental. Simply stated you can't detect a signal transient with good certainty until that transient has actually finished happening, or enough of it to be confident.  One design approach to deal with this for a noise gate (RANE) was to put the audio path through a short delay, that way the un-delayed side chain could have a few mSec of look ahead, more than enough time to open a gate and not compromise the leading edge of the transient.

To make a guitar effect that ignores the initial pick, it seems a simple delay in opening the gate should do that. The speed of the side chain is not that much of an issue if you plan to delay the action.

JR 

> I suspect the RF chip is using an alternate approach.

You know the basic plan. It's old-hat in expensive FM radios.

http://www.analog.com/static/imported-files/data_sheets/AD8310.pdf

It's a cascade of six 14dB-gain stages, surely differential-pair, and nine detectors.

Small signals amplify linearly. Slightly larger, the last stage rounds-over like any diff-pair. Even more signal, that last stage clips and no longer contributes to output. Meanwhile the N-1 stage is going roundy. At max level the first stage goes roundy and clips.

Summing the detector on each stage gives a linear-dB output, pretty near. There should be a woggle every 14.3dB, but there's enough roundy-action and overlap that nominal error is well under 1dB from -70dB to +10dB.

I don't know why 9 detectors for 6 stages.

FWIW: the sheet says "...down to 20Hz." I believe it is DC-coupled, but the least DC offset will foil your low-end accuracy, so they don't promote DC use.

FWIW: input Z is 1K, so you won't hang this right off a geetar.

Correction: I said "as good as any junction-log" but I would NOT want this in a voltage-to-pitch converter as in a synthesizer. The joggles will be invisible to instrumentation but maddening to the musician. The junction logger has flaws but it is _smooth_. If your voltage ratios are correct (precision resistors), the pitch ratios can be made correct over several octaves with simple scaling. (Several octaves but not several hours......)

I don't see how this is much use in an attack-killer. Band-limit, peak-detect, differentiate that signal (can just read the current in the peak-detect's diode), trigger a slow-rise VCA.

Thanks for all the great replies. I appreciate your sharing your experience and knowledge.
@JR - I think there are some digital multi-FX pedals that do exactly what you suggest. I'm trying to avoid any A/D and D/A conversion because I want to use a lot of gain, and any delay I can think of is going to add noise or digital grit when the gain ramps up really high.

@JS - I like the idea of closing the gate in the absence of a signal to help avoid artifacts on pick attacks.

@PRR - thanks for the noting of the problem with the behavior of the AD8310 during the overlaps between stages. You had already sold me on dropping the part, but you are right, smooth matters. You mentioned band-limiting the signal - what about boosting the high end going into the full wave rectifier? Would that do the same thing?

To be clearer(or as my thinking is getting clearer), the log detection is so the swell can follow a consistent envelope shape irregardless of which note whereever on the guitar is decaying at whatever it's natural rate happens to be. Perhaps something akin to what the Transient Designer circuitry does.

Sounds like scope and breadboard time <grin>

Thanks!

Sounds like

> smooth matters.

Not for amplitude. Certainly not for amplitude detection.

I think you are over-thinking.