Pre-emphasis/de-emphasis network for PT2399 chips

[Mark Hammer]

If you were trying to keep any possible noise from a PT2399-based effect out of the audio output, and were going to use the traditional Boss-type pre-emphasis/de-emphasis circuit to do so, where would you stick the corner frequency of the pre-emphasis, and how much emphasis would you opt for?

Or does it even matter?

[stm]

I posed this question to myself long ago in the context of reducing noise from a PT2399.  My first reaction when seeing all those PT circuits without pre/de-emphasis networks is that there was room for improvement with regards to noise by using such network, however a deeper thought made me doubt this approach.  I'll explain in the next lines:

A PT delay is based on delta modulation, which we all know is different from the analog sampling process that takes place on MN and SAD chips.  One weakness of delta modulation is that high frequency signals (and specially those with high levels) produce slope overload (meaning the output cannot follow the input), which creates audible and nasty distortion artifacts that cannot be removed by filtering because they are lower in frequency.  So, adding a pre/de emphasis network may filter high frequency noise better, but at the same time it favours the appearance of undesired frequency components related to the slope overload which cannot be filtered.

A deeper thought makes me think that a compander (without pre/de-emphasis) could provide a better alternative in this case.  For one, it will lower the noise floor at the output, and at the same time it may limit high level peaks that may be more likely to produce slope overload.

As a last comment, if you really want to clock your PT chip slower than the datasheet ranges, the only realworld solution I can see is:

a) Filter the input and output signals real hard (reduce the cutoff frequency and increase the order of the filters with respect to the datasheet or typical circuits out there)

b) Consider that both integrating capacitors may need to be increased in size (I think the standard integrating caps are specified as 82nF, sometimes replaced by 100nF).  Think of it this way: slower clock means longer integrating time, which means the integrator may reach saturation more easily when a slope overload ocurrs.

Unfortunately there's no free lunch here; if you want longer delay times you must give up something, whether it is distortion, noise and/or bandwidth.  This quest of getting more than 300 msec from a PT2399 is no different than attempting to mod a boss DM-2 to get 600 msec instead of the stock 300 msec (without installing a second MN3005 of course) and expect decent audio quality.  The 300 msec audio quality is already on the verge of what is considered acceptable (at least for some people).

[Mark Hammer]

Thanks, Sebastian.  Thoughtful as always.

The basis for the question was a discussion with forum member alparent, who works upstairs from me.  He's been trying to finalize an optimal version of a Dimension P chorus, with variable mixing of the dry and individual wet signals, and when we realized there would be an op-amp there, I said "Why not use a dual op-amp, with an input buffer as well".  And when I realized there would be an input and output stage, I thought maybe a pre-emphasis/de-emphasis network (essentially just 4 additional passive components) might help.

Since the unit has an LFO (which a pure delay might not), there is the risk of ticking.  However, we discussed decoupling the planned individual dual op-amps, and that might avoid any risk of audible ticking and any need to introduce more measures for noise reduction.

Alain is planning to do a layout with some pads for the added networks, and replace them with jumpers if the network seems to not add value.  Not having completed any PT2399-based projects myself, I'm naive with respect to how much noise is generated by different "font versions", so I made some suggestions that may well be overkill.  We'll see what happens.

[stm]

Mark, your insight of the intended use is quite valuable.

Now that you mention it is for the Dim P, I understand we are talking about very fast clocks only (and low delay times, maybe in the order of 30 msec).  In this particular case it could be well possible that pre/de-emphasis will not cause slope overload (due to the nearly maximum clock frequency).  I would try something that provides close to 10 dB emphasis.  Most of the pre/de-emphasis action should take place below the input/output filter cutoff frequencies.

For instance, I would start with a network like this: 22k resistor in series, a 10k to GND, and a 10n cap bypassing the 22k for the pre-emphasis.  In this case the de-emphasis network would be 22k series resistor, and then a 10k+10n to GND.  Final tuning can be done by changing both caps to other values, probably in the range from 2.2n to 15n.

[Mark Hammer]

I'm afraid I'm not following you.

Alain and I are starting with the same arrangement that Boss has used for a variety of BBD-based pedals.  So, let's use the CE-2 as a reference ( http://www.hobby-hour.com/electronics/s/boss-ce2-chorus-schematic.php ).  The first inverting op-amp stage after the BJT buffer has a gain of 1x for the mids and lows (47k input, 47k feedback) and a parallel path, using the 6n8 cap and 10k resistor, that provides a gain of 4.7x above around 2.4khz.  The inverting mixer/buffer stage on the output does the inverse, with the 6n8/10k network in the feedback loop providing complementary treble cut.

So, using that as a reference, what are you recommending?

[stm]

The values I gave were for generic passive divider networks.  Looking at the CE-2 reference schem my recommendation for a first try would be: maintain the 47k, maintain the 6n8 caps, replace the 10k with 22k to limit maximum pre-emphasis to 10dB.  Then tune the 6.8nF caps as necessary.

By the way, the CE-2 pre-emphasis achieves a maximum gain of 5.7x rather than 4.7x.  This is because at very high frequencies the 10k resistor is in parallel with the 47k resistor, forming an equivalent 8.25k resistor.  Thus, maximum gain is 47k / 8.25k = 5.7 times.  In a similar way, the maximum boost with the proposed values (47k and 22k) is 3x or 10 dB and not 2.1x as one would infer from dividing 47k by 22k.

[Mark Hammer]

Interesting.  Hadn't really thought of it that way.  Too much focus on the corner frequency, I guess.  Thanks.

Alain, are you reading this?

[merlinb]

If you were trying to keep any possible noise from a PT2399-based effect out of the audio output, and were going to use the traditional Boss-type pre-emphasis/de-emphasis circuit to do so, where would you stick the corner frequency of the pre-emphasis, and how much emphasis would you opt for?

Or does it even matter?

It's all a matter of headroom. PT can take about 3Vp-p before clipping, so if you were happy for a maximum headroom of 500mVp-p, say, then just preamplify everything by x6. That will allow for significant noise reduction at the output. Or, if you wanted closer to 2Vp-p headroom then preamplify by x1.5.

If that sound too extreme then you can think about where you can sacrifice headroom. For example, the guitar doesn't contain much energy above 1.5kHz, so you could use a pre-emphasis filter that boosts eveything above 1.5kHz at a first-order rate, indefinitely. Then deemphasise with a first-order low-pass filter at the output, much like the Ross compressor.

You could combine the two approaches of course, for example:
Preamplify everything by x2, and in addition do first-order boosting above 1kHz, then level off at a max gain of x6.

[alparent]

Man I like it when great minds discuss.

I am reading this.......

Not understanding 90% of it 

but I'm reading!

[CynicalMan]

Why not use the first internal op amp for the pre-emphasis? You could replace the feedback resistor with a lowpass T filter for a treble boost, and then apply passive filtering on the output. Something like this:

[Suicufnoc]

x
A PT delay is based on delta modulation, which we all know is different from the analog sampling process that takes place on MN and SAD chips.  One weakness of delta modulation is that high frequency signals (and specially those with high levels) produce slope overload (meaning the output cannot follow the input), which creates audible and nasty distortion artifacts that cannot be removed by filtering because they are lower in frequency.  So, adding a pre/de emphasis network may filter high frequency noise better, but at the same time it favours the appearance of undesired frequency components related to the slope overload which cannot be filtered.

Does this slop overload apply to lower frequencies too, or just highs?  (i.e. does it make sense to filter unwanted lows before the delay)

[PRR]

> where would you stick the corner frequency of the pre-emphasis

If the signal is speech/music, we "know" the spectrum is flat 100Hz to a couple KHz then falling. Male speech or choir falls before 1KHz. Some organ passages and disco doesn't fall until 3KHz.





(ABBA is not the best example: they had heavy-bass and some other subgenres were more cymbal-happy.)

If the channel is "flat" and can stand same-amplitude at all frequencies, then you may boost the highs going in, shave them coming out, and gain a lot of S/N.

Of course as Merlin says you should _first_ bring your main signal near 100% of the channel capacity.

Pre-emph works in FM radio. It is available in Audio-CD although rarely used. Tape and phonodisk must trade-off bandwidth for play-time and are shy on headroom in the highs. It could work in AM radio if we used better detectors and if we didn't conventionally chop-off not much above where we might pre-emph (a different space/highs tradeoff).

A generous BBD should be flat. But we always end up pushing for more time and the trade-off with the highs.

As Sebastian is saying, a delta modulation encoder works quite differently. It has severe limits how much highs it can pass. In effect, the delta-mod design "knows" that most signals are not max-amplitude at top-frequency, and has taken advantage of that before you come to the table. I suspect no pre-emph is possible without gross splatter.

Same going through an MP3 encoder. MP3 "knows" that music tracks have declining highs, and that highs bloat file-size, so is tuned to these facts. Pre-emph before an MP3 encoder and it will probably be ugly.

However delta-mod can and usually-does feed RAM, and RAM is ultra-cheap these days. It may be that you can have MHz signals with good amplitude, meaning that audio boosted above 3KHz may pass OK.

> does it make sense to filter unwanted lows before the delay

In straight encoding (WAV or BBD), half your capacity is in the top octave. Conversely 0.1% is in your bottom octave. Similar constraints for ALL encodings. Assuming clean audio, cutting bass is pointless.

If your audio is contaminated with record-warp or subsonic studio blower rumble, cut below your lowest note. I have seen "excellent" phono signals with huge 0.55Hz causing periodinc clipping in later stages. I used to work in a concert hall where the <100Hz rumble was as loud as the harpsicord. This is simple cleanliness. It will not affect "hiss" (unless you actually have to turn-down for subsonic clips) nor length per buck.