First cycle distortion - Graham, what is that?

PMA said:
The explanation hereabove is often used in the DIY community. In fact the electrodynamic speaker can be described by the attached schematics. R2 a L2 are the resistance and the inductance of the voice coil in the "braked" state (not moving). R1, L1 and C1 are the components calculated from mechanical side of the speaker to the electrical one. They describe the resonance effect of the speaker. The component values will differ according to the real speaker.

Exactly.

And you don't hear people referring to capacitors or inductors as "AC generators."

The cone's mass and its compliance are energy storage mechanisms just as inductors and capacitors are. And in fact the cone's mass and compliance have their electrical analogues in inductance and capacitance respectively, which is why dynamic loudspeaker drivers are modeled electrically using R, L and C elements.

se
 
PMA said:
So how about the transient response of the digital filter of the CD player, for example? It has rise time (10% - 90%) no shorter than 17us and limited initial dv/dt, far below slew rate of the contemporary amplifiers. And how about analysis of the transients of the musical instruments itself? ;)

To square the circle, other sources (tape, phono) give pretty low slews also. This looks like yet one more solution in search of a problem.
 
On the subject of speaker electrical models.
The speaker has the additional complexity of being coupled to air and of drive units being physically coupled to one another through the speaker box and the air within the speaker box. Speakers are also microphones, so there will be cone movements which are temporally delayed from the original signal and these will create impedance changes as seen by the amplifier.
Also, the speaker cables form transmission lines which can have very big effects on impedance seen by the amp at HF (as Nelson knows well).
Just for completeness.
 
Disabled Account
Joined 2003
subwo1 said:


That is because they aren't, but a speaker nonetheless is. The model which approximates the real thing is interesting, though.


that's the TS model that I referred to in the original thread. Nobody including Graham picked it up, tho, :)

yeah, if you write out the differential equiations describing the electro-mechanical movements of a real speaker, you get the same differential equations describing that RLC network presented by TS.

so as far as the amp is concerned, the rlc network and its equivalent speaker are one and the same.
 
The one and only
Joined 2001
Paid Member
PMA said:

"Don't bother, it was a terrible piece of work. Amazing he got that thesis passed."

Well I thought the part where he quoted a paragraph of mine
was just great. :cool:

Also, I think there is some merit to the weighting of harmonics
in evaluating harmonic distortion. I know I'd rather listen to
1% 2nd or 3rd than 1% 4th and 5th.

"You cannot get accurate spectral analysis results with an FFT of one cycle....That is a horrible window."

You're right. I'd have to see some evidence of how you can
accurately get Spice to model THD on a waveform start that
has to be full of harmonics by itself, much less steady state.
 
Nelson Pass said:
(...quoting jneutron...)"You cannot get accurate spectral analysis results with an FFT of one cycle....That is a horrible window."

You're right. I'd have to see some evidence of how you can
accurately get Spice to model THD on a waveform start that
has to be full of harmonics by itself, much less steady state.

I believe the problem that jneutron is referring to is the so-called "DFT leakage". Let's assume we have a single cycle of a truly periodic waveform. Ignore for the moment that the first cycle of GM's tone burst, when extended to be periodic, will contain discontinuities. That's because of the transient, which makes the initial value of the "cycle" different from the final value. Anyway, the DFT frequency bins are given by:

fbin= m*fs/N

where m=0, 1, 2, ...N-1 and
N is the number of DFT samples and
fs is the sampling frequency.
The fbin are the frequency "bins", that is, where the DFT is telling us the frequency components of the signal are, as opposed to where they really may be.

Suppose we choose the sampling frequency fs to be exactly N times the fundamental frequency of the periodic signal we're sampling. That is, choose:

ffund=fs/N

Reffering to "Understanding Digital Signal Processing" by Lyons, page 72, he says this:

"The DFT produces correct results only when the input data sequence contains energy precisely at the analysis frequencies given in Eq. 3-24, at integral multiples of our fundamental frequency fs/N"

Andy's note: his equation 3-24 is the same as what I've written above for the fbin.

So if we choose our sampling frequency as exactly N times the fundamental frequency of the signal we're sampling (where N is again the number of DFT points), the bins will line up exactly with the frequency components of the signal we're sampling (assuming that signal is truly periodic). If we don't establish this relationship, the signal will appear spread out in all the DFT bins in general.

How do we make this work with SPICE? SPICE wants to pick its own, often non-uniform time steps based on the time rate of change of the signal. But we can specify a minimum time step, and if that minimum time step is way smaller than what SPICE would compute on its own, we'll get uniform time steps. We can choose these time steps, together with the number of points in the FFT to make the sampling frequency fs an exact integer multiple of the fundamental frequency of the sine wave we're applying.

But the typical approach is to do the analysis on the last cycles of the sinusoid, so that any transient will have died out. Note that this is the opposite of what GM suggests. But GM's measurement really indicates the closeness of the first cycle to that of an ideal sinusoid. This isn't really distortion in the nonlinear sense. Note that I don't necessarily agree with this approach. I'm just trying to explain what I think he's doing.
 
How is it exactly that a speaker is an AC generator but an inductor isn't?
An inductor is passive and is only able to store energy in a magnetic field (normally temporary). A speaker is a mechanical system which can actively convert the mechanical energy of a moving voice coil into electricity as the coil cuts through the lines of magnetic force (from a permanent magnet) during this physical motion. When the speaker is producing a sound, it is a motor, when it is moving in response to sound or simply pushed with fingers, say, it is a generator. The modeling with passive components is good in the standard use of the speaker, when driven by an amplifier, and that is what we care about here. ;)
 
subwo1 said:
An inductor is passive and is only able to store energy in a magnetic field (normally temporary).

Last I looked, loudspeaker drivers were passive as well. And they also store energy.

A speaker is a mechanical system which can actively convert the mechanical energy of a moving voice coil into electricity as the coil cuts through the lines of magnetic force (from a permanent magnet) during this physical motion.

Actively convert? What's the active element in a loudspeaker driver?

When the speaker is producing a sound, it is a motor...

How does calling it a motor change its fundamental electrical behavior?

...when it is moving in response to sound or simply pushed with fingers, say, it is a generator.

But we're not talking about it moving in response to sound or being pushed with fingers. We're talking about it being driven by an amplifier.

But if you want to look at it in that context, then would you also call an inductor which is being impinged upon by a time varying magnetic field a "generator" as well?

The modeling with passive components is good in the standard use of the speaker, when driven by an amplifier, and that is what we care about here. ;)

Sure. So if we're talking about it being driven by an amplifier, then where does the "AC generator" come into it?

se
 
OK. Here is the more complex model, acoustical side included. But nothing is changed from the point of view of the amplifier.
 

Attachments

  • a1speaker.gif
    a1speaker.gif
    6.3 KB · Views: 500
Oh, you actually want to measure it? Well, there's always a troublemaker in every crowd :).

I'm thinking that you might be able to find a function generator capable of tone bursts with controlled start phase (set to zero) and very low "first cycle distortion", with an external trigger pulse to start the tone burst. Then a computer controlled digitizing scope or waveform digitizer, whose sampling was started by the same pulse that triggered the tone burst, might be used. The digitized data might be sent to a computer over the GPIB bus. The computer program that controls all this could also compute the FFT of the incoming samples and display them on a graph.

I don't really know if a suitable tone burst generator exists, nor have I tried to figure out how many bits of resolution you'd need for the sampling. I'm just dreaming here. But I did work for about 10 years in computer controlled RF/microwave test equipment, so it's not total BS, only partial.:)
 
subwo1 said:

An inductor is passive and is only able to store energy in a magnetic field (normally temporary). A speaker is a mechanical system which can actively convert the mechanical energy of a moving voice coil into electricity as the coil cuts through the lines of magnetic force (from a permanent magnet) during this physical motion. When the speaker is producing a sound, it is a motor, when it is moving in response to sound or simply pushed with fingers, say, it is a generator. The modeling with passive components is good in the standard use of the speaker, when driven by an amplifier, and that is what we care about here. ;)

Oh no. The resonant circuit in the speaker electric diagram is really the what you are speaking about (exchange of the energy).
 
The problem with this kind of distortion has nothing to do with either:

Feedback
driver Back EMF
FFT behaviour

It is simply generated by expecting the wrong outcome of the simulation. A simple lowpass would generate the same kind of distortion with the same type of signal.
The signal in question is NOT a SINUSOID but the multiplication of a Sinusoid by a step function. Every lowpass will definitely distort such a signal, like every filter would distort a dirac pulse !
The signal in question and the Dirac pulse have something else in common: The do NOT EXIST IN THE REAL WORLD !

Regards

Charles

/wearing his flame-resistant suit !;)