matching JFETS ?

[Mark Hammer]

But the resistances actually don't have to match.  Consider photocell-based phasers.  VERY little chance the resistances match.  But they sound great, just the same.  Why?  Because the LDRs all respond to light variation at all points in the sweep cycle.

The effect depends on the cumulative phase shift across the stages.  It doesn't matter that much *where* in the spectrum any single stage produces phase shift (within reason), so long as where the shift occurs is moving.  What one tries to avoid is any single stage choking in one spot for part of the sweep, while the others change.

[Rob Strand]

But the resistances actually don't have to match.  Consider photocell-based phasers.  VERY little chance the resistances match.  But they sound great, just the same.  Why?  Because the LDRs all respond to light variation at all points in the sweep cycle. 

They don't need to match in terms of fine tolerance but they have at at least be in the ball park.   Despite LDR's poor tolerance they do seem to fall in that practical ballpark.

The "off point" issue you get with JFETs is largely eliminated with LDR's.    One advantage LDRs have over JFETs is a very wide range of resistance adjustment.   In theory JFETs can act as a high resistance but the Vgs_off mismatch puts a practical limit on the upper resistance.   

For example,  suppose you have two JFETS matched to Vgs_off within 50mV.   Both are supplied from a common gate supply, as is done in a phaser,
  Q1:  Vgs_off = 1V, and
  Q2:  Vgs_off = 1.05V
  Assume rsd_on is perfectly matched at rds_on = 250ohm (yfs0 = 4000 uS)

Suppose we bias both JFETs at 50mV above the lowest Vgs_off of the two JFETs
ie. Vgs = 1V-0.05V = 0.95V.

The resistances each JFETs are,
rds1 =  250 / (1-0.95/1) = 20 * 250 = 5000 ohms
rds2 =  250 / (1-0.95/1.05) = 10.5 * 250 = 2625 ohms

So despite what at first seemed like a good match, the effect of even a small Vgs_off mismatch produces a very large tolerance at the high resistance end.

As the Vgs bias point approaches zero the resistances will both approach the same value of 250ohms.

As you can see the practical range of resistance adjustment is going to be in the 10:1 to 20:1 range.   For a compressor the practical range is higher because we don't have to match two JFETs.

[Links for JFET resistances:
http://educypedia.karadimov.info/library/322Lecture31.pdf

page 12 of,
http://www-personal.engin.umd.umich.edu/~jwvm/ece414/PowerPoint/JFET_Overview.ppt
]

[armdnrdy]

But the resistances actually don't have to match.  Consider photocell-based phasers.  VERY little chance the resistances match.  But they sound great, just the same. 

I hate to disagree with you Mark but, to me...it seems logical that the resistance of the vactrols/JFETs be in a similar range rather than have one move from 2K-100K and the next from 30K-300K.

It seems that matched resistances would have a smoother, more linear sound.
If the resistances were are grossly unmatched, that would be similar to the VGS off being unmatched.

I have performed testing with R.G.s matcher, and the ROG matcher to check the min/max resistances of matched JFETs while connected to a phaser LFO.
I have found that the resistances are similar as a general rule but, in the same "batch" of matched JFETs there will be a few that are "way out" from the group.

Here is one of the factory component pages for the MXR Phase 100. Under "Special Note" it shows that MXR did match the vactrols for this phaser.

[Mark Hammer]

I think our point of disagreement lies in what constitutes "matching".  My own original purpose was to allay any fears some folks might have that their JFETs weren't matched enough (i.e., on multiple parameters).  It matters that they all sweep up and down together (which vgs-matching and proper biasing will take care of), and are - as Rob said - "in the ball park". But more than that is gilding the lily.

I urge people to keep in mind that it is always the sum total of phase-shift at any given frequency, across all stages, that is the objective.  If, at the lowest point of sweep, one stage provides 87 degrees of phase shift at 150hz, another provides 86, a third provides 90, and a 4th also provides 90, then we have 353 degrees of phase shift at 150hz, and don't hit the full 360 until a little higher up in the spectrum.

I might point out that the range of sweep is also a function of the fixed resistor placed in parallel with the JFET, and the cap value.  Getting a reasonably good match on JFETs simply means one doesn't have to monkey around with either of those other components.  But, in principle, as an example, if the zone where a given stage is applying maximum phase shift (as a function of the JFET parameters) swings between 1.3khz and 9khz, then that range could be dropped to 325hz-2.25khz by quadrupling the value of the cap, and even a little further by increasing the 22k resistor to 24k or even 27k.

The MXR note is interesting.  I just wish we knew more about what was being used as a benchmark for "matched", and what the tolerances were for that part at that time in production history.  Unfortunately, it's hard to peek inside sealed optoisolators! 

[tubegeek]

Just want to say a thank you to armdnrdy, Rob Strand and Mark Hammer for your detailed explanation. I happen to be reading through my new copy of Horowitz & Hill and I'm in the JFET chapter lately. Between the three of you and Messrs. H&H, I feel like I am understanding what is going on here at a much better level than ever before.

Of course that's a mistaken impression, but I'm enjoying it so let me have my moment, OK?

[italianguy63]

Just want to say a thank you to armdnrdy, Rob Strand and Mark Hammer for your detailed explanation. I happen to reading through my new copy of Horowitz & Hill and I'm in the JFET chapter lately. Between the three of you and Messrs. H&H, I feel like I am understanding what is going on here at a much better level than ever before.

Of course that's a mistaken impression, but I'm enjoying it so let me have my moment, OK?

Hahahahhaa.. I am just hearing Charlie Brown's teacher... whaaaa whaaaa whaa wa wa...   

Way over my comprehension.

[Rob Strand]

An interesting perspective on how far things can be mismatched is to look at the large capacitor ratio used on the univibe phaser.

[Mark Hammer]

....and that is intended to result in broad and shallow dips, rather than narrow and focussed notches, by staggering where the maximal phase-shift occurs.

I'd be curious as hell to find out how in the deuce they stumbled onto that one.  For me, it's right up there with the Worcestershire sauce recipe.  I mean, it tastes/sounds good and all that, but just what possessed you to think of combining those ingredients?

[R.G.]

There was a paper presented to the 1971/2/3/4? JAES on the spacing of RC time constants in phasers and their audible effects. I can only recall the conclusion that identical RC sections produced the more pleasing effect, better than staggered sections. But I can't recall the metric they used for "better". I keep intending to find that article and read it again, but I keep refusing to pay for it or join the AES to do it.

Another thing comes to mind. Staggered phase shift stages in broadband-quadrature setups. Phasers don't do aligned phase shift differences, of course, but I suspect that there's some good work there for understanding.

[Mark Hammer]

So, we've seen the combining of fixed and swept stages, and we've seen the combining of staggered/non-identical swept stages.

Have we ever seen any cases where identical stages are used, but not swept in identical manner (e.g., some swept only half as wide)?  And would there be any musical value in doing so?

So, as an example, let's feed the gates of two Phase 90 stages from the LFO with one current-limiting resistance, and feed the other two stages with a different resistance.

[amptramp]

There are fixed four-stage allpass filters of the sort seen in a Phase 90 or Univibe used for voice broadcasting.  The idea there is that different frequencies will show amplitude peaks at different times, permitting more signal amplitude without overmodulating the carrier.  These filters usually have dissimilar sections like the Univibe.  This may be an interesting effect - will the music sound louder at a live gig with the phaser connected but not necessarily sweeping?

[R.G.]

Have we ever seen any cases where identical stages are used, but not swept in identical manner (e.g., some swept only half as wide)?  And would there be any musical value in doing so?

You may remember some years ago you and I exchanged some notes on non-synchronous moving notches. It was in the context of emulating the varying resonances/anti-resonances of a sitar bridge, but we talked about phasers as one means to that end.

Psychoacoustically, the ear detects the moving notches of flanging as a detection of motion, and phasers seem to hitch onto a bit of that. Fixed notches are hard to detect, but fixed resonances impart a woody, resonant (well, duh!  ) note, while moving resonances get vocal in some instances. I would guess that musically, non-synchronous notches and/or resonances would hitch onto one or more of these as they moved. It's worth looking into.

Notice that the phase change produced may not even be audible on its own. A moving phase change imitates motion as Doppler effect, but many of the effects of phase change are most easily heard by their notches and peaks. A phaser, with dry mix, is mostly a peaks/notches device, not a vibrato.

So, as an example, let's feed the gates of two Phase 90 stages from the LFO with one current-limiting resistance, and feed the other two stages with a different resistance.

IIRC, that was one of the possibilities I suggested, feeding different pairs of stages from different modulation sources. It's also possible to do this directly, with something like a twin-T per notch and just wobble the notches around directly. Probably about the same number of components, but one might be easier to mess with than the other. The Twin T cell lends itself notches in the forward path, peaks in the feedback path, so there may be some elegance there.

I would think that the synth guys would have plumbed this well long ago, but I don't remember reading anything about it. Anyone else?

[Mark Hammer]

Let's distinguish between differential modulation of stages , in which each cluster is combined with dry signal to produce one or more notches, and that scenario where the summed amount of phase shift prior to mixing is jerked around by feeding some JFET gates a different amount of a common LFO than other gates.

The first case is essentially individually-controlled notches (and any accompanying resonances), in a way that would be similar to a couple of people playing with the individual centre-frequency controls on a multi-band parametric equalizer.  The second case is....hell, I have no idea whatsoever! 

[tubegeek]

You may remember some years ago you and I exchanged some notes....

I saw a picture of that meeting: