KNEE

[WGTP]

I see various reference to GE, SI, LED, etc. clippers having "softer" knees than the other.  (Mine are sort of weak these days.)

Is there anything definitive about this?   

[tcobretti]

In reference to compression, knee refers to how aggressively the signal is compressed once it passes the threshold.  So, long story short, I suspect a hard knee clipper would distort more and earlier than a soft knee clipper.

[petemoore]

  Si's have a harder knee than LED"s or Ge's
 Ge's and LED's represent the ends of the spectrum of clipping thresholds [lowest/highest, without 2X'ing them], and sound different enough from one another in an easy to hear way, hard to word way.
 For clipping applications discussion, because the Opamp characteristics play a role [because LED's may let the opamp have more influence on the clipping sound by how fast the voltage approaches and goes over the clip threshold], how the diodes influence the clipping response can be 'circuit specific'...you asked for 'definitive' information, perhaps choosing a particular circuit for discussion, in which the opamp rails are widened [say 12v or 18v supply] so the diodes/opamp relationships can be further separated.
 Sorry for the 'half-blind' suggestion, perhaps someone can chime in and help us sort out how much the various circuit diodes knees tested and signal preboost need to be separated for diode knee only discussion, or if that's possible.
 Such as in the case of a DIST+, knee's and threshold being more interactive with the opamp characteristics.
 Perhaps the choice of circuit matters in application, discussion of the Knee may not be as exactly applicable from one circuit to the next?
 Anyway, for discussion purposes only, I ask if it would be a good lab technique to fix the circuit in which the diode tests are conducted.
 

[Mark Hammer]

I just finished responding to a posting on just this topic on Ampage.

For purposes of initiating debate (but also because I believe it), I'm going to say that the impact of different "knees" will be minimal if at all audible.  Why?  Because knee becomes important and consequential in determining the audible changes to the harmonic spectrum when you feed it a single steady state/level (as in oscillator) tone.  The moment you feed it a signal as wildly differentiated in its harmonic complexity and pitch stability, as varied in its amplitude and envelope characteristics, from anywhere between 1 to 6 and 7 strings, hearing those knee-related changes is a bit like noticing how one person in a crowd of 10,000 went for a different colour of socks that day.  Add in all the filtering that accompanies most clipping devices, and now you're up to noticing if one person in a crowd of 25,000 has a few more grey hairs than they used to...last week.  That person may well HAVE more grey hairs, and the person may HAVE changed their usual sock colour, but will you actually be able to reliably detect it?  Will it stick out in any way?  Not likely.

For my part, the long and the short of it is that there very likely ARE measurable consequences of such component differences, but being measurable is not the same as being audible in a real-world context, just as being audible is not the same thing as being reliably identifiable.  For instance, we KNOW that GE and SI diodes have very different forward voltage properties as well as other ones.  We are certainly able to notice when a switch has selected between the one or the other type, if nothing else has changed about the circuit, because the large volume change and differential amount of clipping at the same input signal level is a dead giveaway.  Would we be able to reliably recognize germanium from silicon clipping if the input signal levels were adjusted to be equivalently close to the clipping threshold and the gain recovery was adjusted to provide identical volume/listening levels?  I doubt it.  Someone certainly *could* learn to differentiate subtleties, the way that, say, North American Caucasians have a marginal ability to differentiate between Korean and Japanese faces in head and shoulders pictures.  But no one would be able to just walk in and instantly say "Oh yeah, that's a germanium clipping device" the way they can easily pick out a wah wah against a non-wahed guitar. 

Measurability, audibility, and spontaneous identifiability are three separate parameters.  If people can instantly recognize something, then clearly it is audible, and if so it is measurable in some manner (e.g., recognizing that something is "distorted").  But the same sort of guarantee doesn't work going in the other direction.  Measurability is no guarantee of audibility or identifiability.  Clinical researchers see this all the time.  Can you measure a difference between treatment and no treatment?  Yes.  Has it made any subjective or consequential difference in the well-being of the patient/client?  No.

[Gilles C]

And clipping is influenced by the surround circuit, or how it is used. As in this article.

http://sound.westhost.com/articles/soft-clip.htm

Gilles

[Peter Snowberg]

http://sound.westhost.com/articles/soft-clip.htm
http://technology.niagarac.on.ca/courses/etec1120/Files/Unit11.pdf

hehe, Gilles beat me to the first link. 

[Khas Evets]

What Mark said goes against my limited experiments. For example two Ge in series compared to one Si have roughly the same voltage drop, but sound very different. I always thought the knee was responsible for this. I've even heard people around here compare different Si diodes and claim to hear the difference. I've never been able to hear the difference between Si diodes.

[Mark Hammer]

I'm not going to cling like a wolverine to my claims against all reason or evidence.  I will say, however, that I've found considerable variation in forward voltage within a given diode number/type, such that 2 1N34s in series still yielded only about 2/3 the voltage drop of a common 1N914.  GE types have ranged from 180-240mv in my experience, and SI from just under 500 to over 650mv.  So, you MAY be right, but I'm disinclined to start rumours about the identifiable audible consequences of different knees unless I know for sure I'm dealing with comparable test conditions.

Again, there may well BE a difference and an audible effect.  The problem is that normal conditions of use don't permit that difference to be reliably isolated and identified.  Swapping out any two unselected/unmeasured GE in a pedal for any unselected two SI or unselected LED changes not only the knee but the clipping threshold, the relative symmetry, the output level, the driving of any subsequent stages in the pedal, etc.

The other thing I don't know enough about to hedge a bet one way or the other is the real speed of these knees.  Keep in mind that conventional diodes are used for and called "switching diodes", so I'm guessing that the knee may look substantially different on one of those nice scope pictures, but in reality may well reflect a difference that only shows up in the above 100khz realm.  I'm not saying it does or doesn't, just that I have not been presented with any evidence that such an electronic difference would be relevant to that oh-so-critical band between, say, 200hz and 1000hz where all those fundamentals live.

I think we all need (myself included) some hard-nosed realism added to the mix here, just to sort out what is and isn't feasible or relevant.  For instance, it may well be that the clipping knee of some diodes is potentially influential for content above 2khz, where rise times are 125usec (or 250usec if you start from the bottom), but much less influential below that because the transition to clip is so fast relative to rise time that you wouldn't hear it.  So, we might ALL be right, just not all of the time or in all conditions.

[Processaurus]

Technically, if two different types of diodes have the same knee, but different turn on voltages, wouldn't the relative knee (the percentage of the amplitude of your waveform that's getting into the soft clipping area) be more gradual for the diode with the lower turn on voltage?  Ie the signal spends more of its cycle being soft clipped, relatively?

I've kinda wondered about the audible effects of different clipping diodes when I tried a bunch in my rat clone, though.  Hard to say because louder almost always sounds better.

[Ge_Whiz]

One of the most interesting things about the diode 'knee' is that... it doesn't exist.

The so-called 'knee' is only apparent when you look at the diode current vs. voltage curve over a moderately large current range. The curve appears to run from almost horizontal (reverse-biased) to constant slope at relatively high forward currents. Over what seems to be a small range of current/voltage, the slope seems to change relatively quickly from one to the other. In fact, this is purely a visual artifact - largely based on the scaling problem that typical reverse currents are in the microamps, whereas typical forward ones are three or four orders of magnitude higher. Furthermore, this behaviour, continuous through both reverse- and forward-biased regions, can be described by one exponential function with only two parameters - the reverse saturation current, Is, and the 'ideality factor', n. Throughout the diode response curve, this function is smoothly continous, and the suggestion that there is some kind of sudden or even gentle discontinuity, as suggested by a 'knee', is erroneous. Furthermore, virtually all diodes, whatever they are made of, have similar values of n - usually between 1.1 and 1.4. If you were to apply a DC bias voltage to a silicon diode, so that the signal voltage needed to produce, say, 0.1 mA forward current (enter any value you like) were about 0.3 V instead of about 0.7 V (i.e. about 0.4V 'shift' or bias), you would not be able to distinguish the behaviour or sound of the silicon diode from a germanium one. In fact, such a bias control would enable you to 'warp' through a range of typical diode behaviours. Tricky to do for anti-parallel pairs, though...

[WGTP]

Well, that is sort of the point of my post.  Good job guys.  It sounds like we don't know for sure.

Many of us have sat at our breadboards and tried a variety of diode combos GE, SI, LED, Jfet, Mosfet, Transistors, mystical crystals, etc. and hear the differences, by as Mark pointed out, it's difficult to isolate only the "knee" as the variable and not the clipping thresholds, etc.

(Mark, I also like your comments about hearing and not being able to measure [yet] and measuring but not being able to hear.  I think the concept of "if you can't measure it you can't hear it" is sort of arrogant on our part.  You couldn't measure any of it 100 years ago)

It seems to me there is a certain amount of Folklore about this subject.  As always, things may work on way in one circuit and another way in a different circuit.  Feedback loop clipping vs. diode to ground clipping being one major variation.  Still, from my experiments, I associate GE's with a sort of fuzzy sound, but that is usually with a very low clipping threshold.

It may be that all we need is a couple of GE's tied to a 10K pot to vary the clipping threshold.

What I was hoping for was some graphs or visual representations of measurements of clippers showing the "knee" at various frequencies and in various applications, although it may or may not be something we can hear.

http://www.elixant.com/~stompbox/smfforum/index.php?topic=38581.0

I have been using Mosfets as clippers (in the feedback loop to minimize op amp distortion) and they do "seem" to have a darker and more organic sound to them, and each one sounds slightly different.  I guess I need to go back now and compare them to some of the old favorites and see if my 13 year old test subjects can tell the difference.  

[Gilles C]

If one of the differences between the diodes is in the harmonics content of the resultant signal, that could be the reason why it is hard to hear a difference between the diodes used in a circuit.

Another interesting article about the use of diodes.

http://www.headwize.com/projects/limiter_prj.htm



Gilles

[Ge_Whiz]

...they do "seem" to have a darker and more organic sound to them...

No personal criticism intended, but it's statements like that one that make a rational discussion of stompbox design difficult. We're back in 'mojo' territory. Sorry if that comes over too harsh.

Incidentally, adding a 10k pot in series does not make a variable-threshold diode. A germanium diode makes a more fuzzy sound just because it clips harder for the same input signal. Stray capacitance may also produce a filtering effect, neither of which has anything to do with 'knees'. Harder clipping causes a greater proportion of dissonant harmonics to appear, which at some level, most people find unpleasant. Hence, most people prefer the sound of clipping caused by diodes with a higher Vf, which, all other things being equivalent (please note), is the same as presenting a lower Vf diode such as germanium with a lower-level version of the same signal. Note that I am not saying "all diodes sound the same" - they present different capacitances and impedances to the signal path, and thus sound subtly different - but because of frequency response, not knees.

[petemoore]

As always, things may work on way in one circuit and another way in a different circuit.
 May not be a pertinent addition to the subcontexts...
 Most of the diodes used use a cap to 'round' the clipping...does this influence the way the knees of diodes work?
 

[Ge_Whiz]

Gilles, nice article, except for:

"A diode has infinite resistance until the voltage across it is high enough to forward bias it (typically 0.7V for a silicon diode) at which point, current flows."

Please, don't anybody take this as being the truth. It's a horrible oversimplification, but adequate for child-friendly articles.

Pete: no, capacitors do not affect the (non-existent) 'knee' - or, more accurately, the behaviour of the diode itself.

[WGTP]

No problem, that is why I use "seems" alot and the type of discussion I wanted. 

There is also a problem with the GE and 10K pot idea, in that your adding resistance in series with the diodes.  Back to the 6-way switch.  I noted when looking at Mosfets at the Supertex site, that they have different capacitance values, etc. that could enter into the mix. 

Some folks use the 4001's instead of 4818's.  Some like Green LED's instead of Red. 

It's starting to look like the "Knee" deal is less significant than other factors that change when clippes are changed...