Transient Response Testing

[Chris Lockwood]

Sorry if I was misleading with my initial posts.

Some of it was wrong- especially the bit where I said "voltage across the voice-coil is proportional to velocity". This does not apply in general. The transfer function is anything but constant

Current on the other hand is closely proportional to acceleration (=>SPL) in the midband, controlled by compliance.

My brain was a bit rusty on this and I was a bit quick to respond...

[MBK]

Pardon if I miss the point - but isn't a square wave composed of (ideally) infinite high order harmonics? And isn't therefore a perfect step response limited by the inherent *LP* behavior of the speaker? And shouldn't the initial poster rather be testing using modulated tone bursts a la www.linkwitzlab.com ? Linkwitz has a lot of burst test info on his site (if it's still there after the recent pruning). His graphs look similar to the ones presented here.

MBK

[Svante]

Originally posted by Chris Lockwood
Svante,

We are just looking at the same problem from different perspectives, with different implementation methods.

We all agree that we are trying to control the cone accelleration. Loop stability is the big issue here.

Yes, we are right!

For me the issue was the transfer function voltage -> sound pressure. But of course I would need stability as well, otherwise the transfer function would be pretty meaningless.

[CeramicMan]

Originally posted by MBK
Pardon if I miss the point - but isn't a square wave composed of (ideally) infinite high order harmonics? And isn't therefore a perfect step response limited by the inherent *LP* behavior of the speaker? And shouldn't the initial poster rather be testing using modulated tone bursts a la www.linkwitzlab.com ? Linkwitz has a lot of burst test info on his site (if it's still there after the recent pruning). His graphs look similar to the ones presented here.

MBK

A step input has a frequency sprectrum that's sloped at 6dB/octave, and infinite harmonics like you said. By definition, an impulse response can be integrated to produce a step response, as a step response can be differentiated to form an impulse response. A modulated tone burst only tests a small range of frequencies with variable amplitudes, which is arguably less useful than testing all frequencies at once.

A step input is also very intuitive because even with inadequate analysis tools, a person can still learn a lot directly from the step "response" of the speaker. If the speaker is under-damped at a certain frequency, it means that at that frequency the total Q-factor is more than 0.707 and there will be ringing, and if it's less than 0.707 there won't be ringing. It's downright obvious that there's ringing if something that should be a nice curve has little ripples on it instead.

The downside of impulse response testing is that the amplitude has to be very high at high frequencies before anything measureable is recorded at low frequencies. Therefore, a step function can be used, which is then differentiated so the result is similar but with lower overall distortion.

The result is a graph with a little seismic ripple on it, and an FFT can be performed on the data. If all of the data is converted at once, a frequency response graph can be produced. If only a small snapshot of data is converted, only the frequency content at a particular section of time is known, and multiple snapshots of time are converted to create waterfall plots.

The speaker in question can be perfected if accurate sine and cosine frequency response data is gathered, and an FIR filter is generated that accurately flattens the frequency response.

Lech

[Chris Lockwood]

MBK,

The problem is to create a functional MFB system. The MFB system is a feedback control system.

To solve this problem, we need a characterisation of the components in the system. If we assume that the system is approximately linear, and add some margin in our design to allow for the grossest of errors, we only need the frequency response of the system.

Specifically, we need the frequency response (magnitude and phase) of the accelerometer output with reference to the woofer input voltage.

The best way to measure this is the easiest way we can with the tools we have available, and assuming the feedback transducer is of high quality, we dont need to be too precise with our measurements. Sine wave continuous testing is good enough, using a signal generator and dual channel scope. We can measure the amplitude and phase for, say, 20 different frequencies, and join the dots on the graph.

Plenty of other methods are available, including impulse testing, MLS, and tone bursts- all have their advantages and disadvantages. USe what you have on hand. Remember, it only has to be done once

The plots alone will show a complex relationship. We use accurate models to identify the most predictable factors in the system. Measurements will give us the important variables (such as gain of the accelerometer).

If we already know the woofer mechanical characteristics (Bl, R, Mms, Cms, Rms), we plug it into the model and it should come up pretty close to what we measured. The remaining variable will be the accelerometer gain. This can be calculated by taking the mean of the magnitude response from measurements, and comparing to that of the predicted response.

We then take the model, and choose (or design from scratch) a control system configuration that suits it. Normally we might use current + velocity feedback, filtered acceleration feedback, or some other kind of arrangement...

The measurements will give us the *real* phase response, which tells us how much loop gain we can safely apply.

The feedback will then compensate for the deficiencies in the driver, AND all the approximations we made in our model. Remember, we left some margin in the design (normally the phase margin) to cater for the most gross of errors which might otherwise cause system instability.

Existing models for loudspeakers are very accurate in terms of predicted frequency response.

Once we have determined our control system layout, we then need to work out the maximum amount of feedback to apply before the system goes unstable. The easiest way is actually to build the system, and then do some final testing/adjustment.

With the loop open, we get a dual channel cro, and feed a signal generator into the system. The feedback point is compared to the input of the amplifier. (We need to include the accelerometer integrator or low-pass filter in this).

Wind the frequency up until the phase reaches approximately 135 degrees. Adjust the level of feedback until the input signal level matches the open loop feedback level.

Close the loop.

Now we have a nice phase margin and a low-distortion woofer!

Just add water...

[cm961]

Hi guys,

I finally got some positive results. I've spent a ton of hours to get this far however there are still many problems. I haven't been able to get it working with an integrator. Right now the accelerometer is being passed through a 2-pole low pass filter (sallen-key). I'm still having major phase issues. The bitmap attaches shows that the bandwidth is extended. This is really neat. However, the impulse response is worse than without the feedback and I can tell that the group delay is pretty bad.

Pete

[Svante]

Originally posted by cm961
Hi guys,

I finally got some positive results. I've spent a ton of hours to get this far however there are still many problems. I haven't been able to get it working with an integrator. Right now the accelerometer is being passed through a 2-pole low pass filter (sallen-key). I'm still having major phase issues. The bitmap attaches shows that the bandwidth is extended. This is really neat. However, the impulse response is worse than without the feedback and I can tell that the group delay is pretty bad.

Pete

For those of us who has been away from regulators for a while, could you post the structure (diagram) of your regulator?

[synergy]

ok i know i'm late on the post and i've seen somebody else mention this but you maybe having problems with your signals being lined up correctly

you've got one side going straight into your soundcard then the other going through an amp throught the speaker and then across the air gap to the mic and then into your soundcard

you need to delay the one that goes straight to your sound card and line them up first to get rid of the air gap (thats the one that delays the most) before you'll get any accurate results

try this on a fully working 30 day free demo

drawbacks on the demo - you cant save anything and it is only 30 days it writes something somewhere buried deep in your registery that prevents you reinstalling but then it is a $695 program used for setting up stadiums and the oscars of all things so you cant blame them