Half-wave rectifiers for more headroom...?

[MrStab]

The thread title is a bit condensed, but here's a thought i need someone to shoot down in flames because it's been bugging me for a while (expect some serious knowledge gaps):

If you were to rig up an inverter and a coupla active half-wave rectifiers in such a way, and sync all relevant clocks, would it be feasible to split a signal for separate processing via. chips with limited headroom, such as the PT2399 or BBDs? so one IC processes the top half and another the bottom?

considerations:

• Usability of active rectifiers? AFAIK they exist to avoid diode-drops, but no free lunch...
• Likelihood of successful clock timing? Though that's chip-specific and less the key focus here
• Other tolerance issues, relating to the wave splitting & recombination in particular?
• Recombine by inverting then summing? Difference amplifier?
• Stuff i forgot but may add in a subsequent edit?

I have no free breadboards or time to pursue this at the moment, so i thought i'd ask you guys to see if the basic theory checks out. i can't seem to find anything specifically relating to this, save a post on the forum about splitting for a fuzz and various articles about octavers. i've also considered using relative voltage to my advantage (10-5V for one IC and 5-0V for the other), but figured that could cause a whole other set of problems. Translate to bipolar reference points as needed.

Someone must've tried this already, and it seems like such a no-brainer that it probably didn't work. i need someone to promptly kill the idea (or entertain the slight chance it may work).

Discuss!

[ashcat_lt]

I've got no information on the practicalities of this, but I'm interested, so I'm posting to get it to pop up in my "new replies..." list.

[R.G.]

This is one of those ideas where the devil is in the details.

It's taken about a century of work in electronics to get the bright idea of having one power device amplify the positive half cycle of a signal and another the negative side, then recombine them so that there is vanishing small distortion caused by the recombination. (That is, Class AB and B amplifiers.)

There are ways. Yes, active rectifiers will be needed, as no good way exists to get over the loss of the signal voltage equal to the diode forward voltage when crossing from one polarity to another. The human ear is excruciatingly sensitive to disturbances right around zero, we humans have found. Yes, clock synchronization will be necessary. You'll chase yourself silly trying to get the audio to recombine right if the clocks are not synched with great accuracy. The PT2399 does not have the clock brought out to a pin, so this gets really, really hard to do with that chip.    

I have this ugly suspicion that using a compander chip to compress the input and expand the output could add more useful headroom than using two different delays processing alternating half cycles. Probably smaller, cheaper, and better eventual performance, as companding does other good things for you.

[Transmogrifox]

I think we glean that the short answer is the concept is solid.  There can be no filtering applied between splitting and recombination and the clocks must be perfectly synchronized.  The details will make it challenging.

I have this ugly suspicion that using a compander chip to compress the input and expand the output could add more useful headroom than using two different delays processing alternating half cycles. Probably smaller, cheaper, and better eventual performance, as companding does other good things for you.

The same ugly suspicion also whacked me over the head with its lumpy cudgel. 

The technique you described doubles the headroom (if you can pull it off).

The compressor even with a 2:1 ratio and threshold at, say, nearly - 40 dB, will give you approximately -20 dB output signal for a 0 dB input signal.  That translates to a factor of 10x headroom.

You might come out better splitting bands than splitting polarities (high pass in one path, low pass in the other).

[MrStab]

hmm, so companding is a vastly more straightforward solution, both in terms of execution and result. That probably explains why wave-splitting isn't in a million effects already. What about wave-splitting AND companding?!

it's been a while since i played with PT2399, but i'm starting to faintly remember some posts about the difficulty of clock sync. i've only worked with dual-PT reverb projects with independent timing. Probably not the best example to go with, in any case.

thanks for the insight, guys. maybe there's some application of this concept that doesn't involve clocked components - i'm struggling to think of any, but maybe someone else will.

[Transmogrifox]

What about wave-splitting AND companding?!

It would be fun to try even if it doesn't buy you much in the hi-fi realm.  If I had time I would do it just to see if I could

[MrStab]

Class-G companded reverb here we come!

[hymenoptera]

Wouldn't twice the noise be a consideration, too? When you mix both chip's outputs back together you'll still have the noise to deal with.

[crane]

Please someone correct me if I'm wrong -
Rectification is a non linear operation - adds more harmonic components - this means that even if you did everything correctly each of your both systems should be able to handle much wider bandwidth than original signal has. My two cents to this.

[MrStab]

perhaps these caveats could shape the direction in which to take this concept? (... i still can't think of one!)

[merlinb]

You don't need to chop the signal up, just make it balanced (two out-of-phase signal paths). This effectively gives you twice the headroom.

[Transmogrifox]

Please someone correct me if I'm wrong -
Rectification is a non linear operation - adds more harmonic components - this means that even if you did everything correctly each of your both systems should be able to handle much wider bandwidth than original signal has. My two cents to this.

This is correct.  If you don't apply any filtering (read band-limiting) or nonlinear operations that work near the zero-crossing, then you will be able to reconstruct it perfectly.  For the case of a delay line, you would want to apply your anti-aliasing filtering before splitting the signal, and then you would want to recombine these before application of reconstruction filtering.

If both clocks in the delay line are synchronized, then you don't have to worry about aliasing on the discontinuous break as long as both sides are combined prior to reconstruction filtering.

(Un)fortunately most A/D chips marketed for audio use include some sort of oversampling, filter, decimation and the D/A converters usually integrate oversampling and reconstruction filtering.  Then the question is whether the amount of band-limiting applied to the discontinuity is audible after reconstruction, or if it's high enough to avoid being audible.

just make it balanced (two out-of-phase signal paths)

merlin is right.  That is equivalent but removes the challenge of precision rectification and recombining the nonlinear halves in a linear manner.

If we're talking about maximizing headroom, I still think the compander is the best bang for buck in my book.  When you split signal paths for parallel processing, clocks have to be perfectly synchronized and all the challenges that go along with that, but at least it doesn't need to be quite as precise when using parallel linear processing.

Even better is using a system in which the digital resolution (or SNR) is more than you need.  Then you can create as much headroom as you need with a resistor divider and then amplify the output back up to your full analog swing.

For example with a PT2399, 16 bit precision A/D, D/A you don't need to do anything more tricky than that for headroom.  For example, if you want 6 dB headroom, your effective bit depth becomes 15 bit. 

PT2399 claims "better than 90 dB SNR" (this hints at 16-bit digital -- 20*log10(1/2^16) = -96 dB).
From here,
http://neunaber.net/blogs/brian-s-notes/13830765-the-analog-myth-chorus
I glean a typical BBD can give you -80 dB SNR, and I will assume this is the best you can get on a pure sine wave constant in amplitude at the full headroom.

So, to get that bad with the PT2399 you can use a resistor divider to achieve 10 dB headroom.  Added to that, you have noise from the compander and usually multiple BBD chips...this is why I think you can have all the headroom you would need on a PT2399 without doing anything tricky.

With a BBD I think you don't have much choice other than companding to get adequate headroom and SNR.

[MrStab]

You don't need to chop the signal up, just make it balanced (two out-of-phase signal paths). This effectively gives you twice the headroom.

wow - kicking myself! i'm really surprised i've neither read that anywhere or just figured it out as an inherent implication of balanced signals. i guess everyone's always banging on about the noise rejection aspect with balancing, and less to the headroom benefit. as annoying as the thought of internal balancing is, it'd probably be a lot less hassle than the rectifier thing. thanks for pointing that out!

i arbitrarily picked the PT2399 as an example, Transmog, but it's good to read further elaboration of why that IC aren't relevant to the issue at hand here (and why BBDs are)

[ashcat_lt]

Remember how we talked about the extra noise we'd get from running two of the delay circuits and mixing them together at the end?  So, I'm wondering if this "balanced" configuration actually helps either.  Common mode noise (clock or power supply, RFI/EMI to an extent) will be suppressed, but I'm thinking a lot of the noise we're adding to each side is going to be essentially random or at least different enough that it doesn't count as common mode anymore. 

At that point, one starts to wonder if a simple resistive divider at the input and a bit more makeup gain on the output wouldn't yield similar or better results as far as dynamic range is concerned.

[Transmogrifox]

At that point, one starts to wonder if a simple resistive divider at the input and a bit more makeup gain on the output wouldn't yield similar or better results as far as dynamic range is concerned.

SNR (signal/noise Ratio).  No matter which way you go about it the topic of headroom is a discussion of a signal's dynamic range and the strained relationship between whether you'd rather tolerate distortion on transients or increase background noise.  Headroom is that margin between the noise floor and clipping threshold...often expressed as dynamic range.

As you have pointed out two channels in parallel will double the noise -- but so will reducing the signal level by 6 dB and then doubling the noise with makeup gain.

You just can't win unless you smoosh the signal's dynamic range so signal is always making maximum use of the channel's dynamic range...or use a channel with a very wide dynamic range (such as a high quality AD/DA & digital processing/transmission in between).

When using a quiet channel, then this question of headroom becomes a moot point.  You can always pre-attenuate the signal to get maximum tolerable signal-noise ratio and post-amplify on the output.  Increasing headroom is a trade-off with signal-noise ratio.  The compander lets you cheat a little bit in this game because you effectively increase the channel dynamic range by means of reducing the signal's dynamic range when it's going through the channel.

[merlinb]

Remember how we talked about the extra noise we'd get from running two of the delay circuits and mixing them together at the end?  So, I'm wondering if this "balanced" configuration actually helps either.  Common mode noise (clock or power supply, RFI/EMI to an extent) will be suppressed, but I'm thinking a lot of the noise we're adding to each side is going to be essentially random or at least different enough that it doesn't count as common mode anymore. 

There would still be a noise advantage because the random noise is uncorrelated, whereas the signal is correlated. In theory, this would give a 3dB SNR improvement over a single signal path operating at the same signal level as one half of the balanced version.

If you attenuate the signal before making it balanced then there is a noise penalty of course, but it may well be smaller than the noise advantage you get by running balanced through noisy chips like PT2399s, so you still win in the end. But if you're only dealing with quiet chips like opamps then it may not be worth balancing.
If you don't attenuate before making it balanced, than you may still get a (small) headroom advantage because many chips don't clip symmetrically. (Though this is probably splitting hairs!)

On the other hand, the wet signal of a delay is often so heavily filtered that mild clipping passes unnoticed, or even add a pleasant tape-saturation effect, in which case the limited headroom of the (5V) PT2399 is not really an issue that needs to be solved...

Many DIY FX circuits are so noisy in other ways (e.g. using the wrong opamps, using too large resistor values) than these 'clever' noise-reducing ideas are wasted effort. The rest of the design needs to be well designed and quiet first, before you ponder companding, quasi-companding, balancing etc.

[PRR]

Another approach (seen on several pedals) is to cut the lows (which overload) without cutting the highs. Pre-Emphasis. At the end, use the complementary EQ to restore flatness. As Merlin notes, this heavy high-cut (more than you would use for simple low-pass) takes the hash off of the clipping so it may go un-noticed.

The original premise: Overall, I'm thinking that more than double the complexity (two processors plus splitter and combiner, plus alignment of the two halves) for less then 6dB improvement, is a poor buy.