Optimising Bass Microphone Servo control

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
The main issue with the new Poweramp section is, that the Amp gain is not 35 dB as in the original B&M implementation, but only 26dB. This would mean that I need 10 dB more gain elsewhere in the loop, so would probably have to change the diff. Amplifier gain accordingly. right?...Thinking a bit further f(I) is already known with 32Hz from the original design and may be OK to stay, so the adder gain and the microphone gain seem to be the real "tuning knobs" ...
Are you keeping the same woofer in the same enclosure? If so, the setting of f(I)=32Hz is probably already close to optimal. If you set it any lower, you may wind up over driving the woofer or running the amplifier into clipping easily if music has bass content < 30Hz. I think it is a good idea to have a variable gain stage just upstream of the power amplifier to be able to adjust loop gain. The open loop response measurement is the easiest way to show what you can increase gain to and retain a stable system.

I posted some measured trends showing open loop and closed loop response for a range of f(I) from 5Hz to 50Hz and open loop gain from 5dB to 20dB. It was using an accelerometer, but your mic based system will have similar trends. Note that the peak of the open loop gain curve will be at the resonance frequency of the closed box woofer, so will be dependent on the woofer/box combination even if the MFB circuit is unchanged.
Commercial motional feedback woofer available sort of
 
... Are you keeping the same woofer in the same enclosure?
Yes! Each of the two drive units per box has an individual, sealed volume.
... setting of f(I)=32Hz is probably already close to optimal...
I agree with that.

The old electronics employed one poweramp for two drivers in parallel and summed the two microphone(!) inputs into one control loop.
I'm planning to break that up and control each driver individually with its own power amp and own mic driven feedback loop. Is that overkill? I'm inclined to say "no", but want to try that out as part of the hobby ;)

... the peak of the open loop gain curve will be at the resonance frequency of the closed box woofer, so will be dependent on the woofer/box combination even if the MFB circuit is unchanged...
That's my understanding as well.

As prework I have done Qtc and fsc measurements of the drivers, finding Qtc of 0.57...0.58 and fsc 60...61Hz consistently over the four boxed drivers.

To make all this a bit more tangible, attached arre a few pictures.

Thanks and Regards,
Winfried
 

Attachments

  • DSCF0680 klein.JPG
    DSCF0680 klein.JPG
    703.8 KB · Views: 207
  • TT Mikro Detail 1.JPG
    TT Mikro Detail 1.JPG
    143 KB · Views: 229
…I'm planning to break that up and control each driver individually with its own power amp and own mic driven feedback loop. Is that overkill? I'm inclined to say "no", but want to try that out as part of the hobby ;)
Thanks for sharing pictures of your project :)
Having separate sensors, amplifiers, and feedback loops for each woofer is the theoretical ideal. But in practice, I have found that the response and non-linear behavior of 2 woofers of the same type are similar enough that there is very little to be gained from using separate feedback loops. But, as you say, this is your hobby and from a theoretical stand point it is not overkill.

BTW, when splitting the single feedback loop into separate feedback loops, be sure to use separate R157 supply resistors for each mic capsule rather than sharing one. Also, increase the value of each R157 to 30K so the first order LP filter (mentioned in post #14) remains at 1kHz. Alternatively, you could leave R157 = 15K and add an additional 4n7 capacitor across each mic capsule.
 
Thanks for your advice, which reminds of another point, though:

When I have the power amps up and running, I'll do a measurement of how much the low/mid driver "crosstalks" into the microphone feedback loop as it's next to the woofers. My suspicion is that the mic adds Woofer and low/mid driver sound pressures, thus giving a "false" (i.e. potentially too high) feedback signal back into the loop, falsely reducing the Woofer level in the overlap region. My thinking (which may be wrong!) is, that it may (at least theoretically) be better to have the mic signal filtered with the same low pass as the woofers' cross over filter, so the low/mid "crosstalk" into the mic is attenuated. Well, as I said, my thinking may be wrong, I am unsure... Any advice or ideas here?


Thanks again,
Winfried
 
When I have the power amps up and running, I'll do a measurement of how much the low/mid driver "crosstalks" into the microphone feedback loop as it's next to the woofers. My suspicion is that the mic adds Woofer and low/mid driver sound pressures, thus giving a "false" (i.e. potentially too high) feedback signal back into the loop....
Yup, that's the problem with mic's, eh.

Actually, far more horrible than you intuit. It not just a generic "negative" feedback. The sounds from other drivers will show up at the mic in every kind of weird phase too.... sometimes howling.*

But not a problem with accelerometers (or if the drivers are far apart).

About your earlier query about one mic for two or more drivers, Bolserst certainly is the authority and some past commercializations have worked that way, I have read. But my gut feel is that the feedback is introduced to make small faults smaller and certainly two drivers must have slightly different small faults. Anyway, it just seems wrong to have a single sensor controlling two systems, which seems to be your gut feel too.

B.
* Actually, "the big elephant in the room" is measurement. Other than detecting gross bad sound (or howling), there's no consensus on how to measure MFB. So even if the mic is picking up erroneous errors, hard to evaluate the extent.
 
Last edited:
Speaking of devices and experimenting, I found I could hardwire a 9v battery to the ACH-01 and it seemed to last long. Is it feasible to work with batteries (which have a whole bunch of convenience advantages and low noise too) with any of these sensors? Is the battery drain bigger when driver is active?

B.
 
… My suspicion is that the mic adds Woofer and low/mid driver sound pressures, thus giving a "false" (i.e. potentially too high) feedback signal back into the loop, falsely reducing the Woofer level in the overlap region.
If the crossover frequency between the woofer and low/mid driver was lower in frequency this might be more of a problem. But since it is at 180Hz, the MFB loop gain will already be < 1, any acoustic pickup from the low/mid will have little effect on the woofer response. Even so, it would be best to minimize its contribution by positioning the mic on the upper woofer as far from the low/mid as possible(as shown in your picture). If you were crossing at 80Hz it would be much more of an issue.

My thinking (which may be wrong!) is, that it may (at least theoretically) be better to have the mic signal filtered with the same low pass as the woofers' cross over filter, so the low/mid "crosstalk" into the mic is attenuated.
I would not recommend this, at least as a starting point. Adding a higher order filter within the feedback loop will compromise phase stability. You want the roll-off slope to be first order as it transitions thru unity gain. The integrator circuit with slope starting at 32Hz already does this. Any additional roll-off slope that is added will contribute additional phase lag which will quickly turn negative feedback(good) into positive feedback(bad).
 
Last edited:
Hi Ben,

thanks for your thoughts an the warnings! Please consoder, though, that I have a working feedback loop at hand which I'm evaluating for optimisations and most effective tuning of the (currently fixed) feedback loop to the individual drivers. No howling was detected so far, but a 6dB drop at the 180Hz cross over acoustical frequency response.

Hello bolserst,

thanks for your instructive comments! I can follow your phase lag vs. stability considerations. The C across the mic. output seems to do what you suggest. So, what I have measured (near field) could well be the price payed for acoustic feedback ... right?


attachment.php



So, in summary, it seems best to stay with two mics. and the Cs across the mic., evaluate two individual loops, though.


To my mind the most prominent motivation for mic. feedback-loop introduction at B&M was to extend the bass range (to ~30 in ma case), other linarisation effects coming as a bonus. Feedback with a mic. was "just" low cost compared to the inductive feedback loop (as used in the higher priced B&M Systems until today). Means: AFB is/was an accepted / calculated cost vs. quality compromise. I actually do see mic. feedback loops in quite well souding Sub-woofer still these days (where "crosstalking" sound sources like mid range drivers are relatively far away...).


Thanks and kind Greetings,
Winfried
 

Attachments

  • F-Gang Summe TT&MT 4n7 klein.png
    F-Gang Summe TT&MT 4n7 klein.png
    18.8 KB · Views: 311
… I can follow your phase lag vs. stability considerations. The C across the mic. output seems to do what you suggest. So, what I have measured (near field) could well be the price payed for acoustic feedback ... right?
I’m not sure exactly what was going on with the measurements in Post#10 where you mention seeing a 6dB shift in woofer response. Changing C should have only affected the response at frequencies higher than 1kHz. I suspect that something else was unintentionally changed at the same time to obtain the results you measured.

To better explain the phase lag concern when trying to roll off HF response of mic to help mitigate inputs from the midrange, see attached set of 4 plots.

Plot (A): This is measured response of a MFB woofer system I built with integrator frequency of 10Hz and open loop gain of 15dB.
Plot (B): Modeling of the MFB system…not a perfect match, but very close.
Plot (C): Modeling if a 500Hz LP filter was added to the feedback loop you can see the open loop response is rolled off, but at the cost of phase lag resulting in reduced phase margin and peaking in the closed loop response. (ie marginally stable)
Plot (D): Modeling if a 200Hz LP filter was added to the feedback loop in an effort to remove the peak, you can see the open loop response is rolled off even more, but the phase lag results in complete phase reversal before open loop gain < 1 and sustained oscillation.
 

Attachments

  • LPfilter_reduced_stability_4plots.png
    LPfilter_reduced_stability_4plots.png
    171.9 KB · Views: 131
Plot (A): This is measured response of a MFB woofer system I built with integrator frequency of 10Hz and open loop gain of 15dB.
Plot (B): Modeling of the MFB system…not a perfect match, but very close.
That's an astonishing accomplishment: you took a speaker (small sealed box?) with an awful 15 dB peak around 60 Hz and using MFB you made it flat 12-300 Hz.*

Sure would be nice to see distortion and pulse results.

Nice to see struggles to apply MFB to north of say, 200 Hz. But why try? While the benefits to woofers are substantial, the benefits even to upper woofers and above can't be anywhere near as perceivable. On the other hand, since I've never seen MFB applied to anything but a woofer, does anybody really know?

B.
* OK, I admit I'm also astonished at the agreement between theory and measurement, but I guess a real pro like bolserst can manage it
 
Last edited:
Harmonics (distortion) can go up to say 100 Hz, so control at those frequencies makes sense for distortion reduction.
For a significant reduction you need a large open loop gain at the harmonics that need to be suppressed. A high open loop gain at 100 Hz means that the gain crossover frequency can be 200 Hz.
 
Last edited:
That's an astonishing accomplishment: you took a speaker (small sealed box?) with an awful 15 dB peak around 60 Hz and using MFB you made it flat 12-300 Hz.
Not quite…remember the open loop response includes not only the woofer response, but any filters included for improving loop stability(like the integrator). The woofer was in a small box and underdamped (DVC running with just one VC) but it only had about a 3dB bump at resonance(Q=1.3) and some HF roll off from VC inductance. (see attached plot with woofer response added). Only one VC was being driven because the other VC was being used to compare VC feedback methods with accelerometer feedback.

Once open loop response(magnitude and phase) has been measured(lets call it H) determining the closed loop response is just a matter of calculating H/(1+H). So, one would hope theory and measurement wouldn’t be too far off. Then, adding in a LP filter(G) to evaluate impact to closed loop stability is easy enough; the closed loop response is just H*G/(1 + H*G).

Sure would be nice to see distortion and pulse results.
Remember, the pulse response and the frequency response are mathematically tied together. No new information, just a different way of looking at the same information. Are you interested in the impulse response? Or tone-bursts? If so, what frequency. For me, the frequency/phase response is a more intuitive way to look at the information.

I did measure distortion, but at that time the equipment I used for distortion measurement was stand-alone and did not have export capability. In any case, with the ACH-01 accelerometer, distortion as measured by microphone was reduced by the expected amount of the open loop gain factor at the harmonic frequency. Example for 30Hz shown in attachment.

Nice to see struggles to apply MFB to north of say, 200 Hz. But why try?
As TBTL mentioned, and I alluded to above, if you are trying to reduce harmonic distortion you need to push open loop gain to as high a frequency as practical. HP and LP filters are used upstream of the MFB circuit to tailor the subwoofer bandwidth as desired.
 

Attachments

  • MFB_Dist_Reduction.png
    MFB_Dist_Reduction.png
    43.4 KB · Views: 186
  • MFB_Woofer_response.png
    MFB_Woofer_response.png
    25.5 KB · Views: 186
Not quite…remember the open loop response includes not only the woofer response...

Once open loop response(magnitude and phase) has been measured(lets call it H) determining the closed loop response is just a matter of calculating H/(1+H). So, one would hope theory and measurement wouldn’t be too far off...

Remember, the pulse response and the frequency response are mathematically tied together. No new information, just a different way of looking at the same information. Are you interested in the impulse response? Or tone-bursts? If so, what frequency. For me, the frequency/phase response is a more intuitive way to look at the information.
You are far too modest. Anybody reading other sub threads would be gobsmacked* by your result.

Granted, in theory nothing in pulses that aren't in FR. There's "theory" and then there's "meaningful communicative display that tells your eye what your ear hears". Theory aside, in practice I sure would like to see before-and-after output traces for say speaker resonance and say, 80 Hz.

Or even better, snap-shots of Dirac pulses perhaps with their RTAs. Or sound recordings of Dirac pulses before-and-after. When I did serious research on MFB, nothing was more convincing than the recordings of "click" versus "clunk".

But what were your results using the second voice coil for feedback?

B.
* don't think I've ever used the word "gobsmacked" before
 
…even better, snap-shots of Dirac pulses perhaps with their RTAs. Or sound recordings of Dirac pulses before-and-after. When I did serious research on MFB, nothing was more convincing than the recordings of "click" versus "clunk".
Since those measurements were calculated from their impulse response, this should be easy enough to post…assuming I didn’t pitch them. I’ll see what I can find this weekend. Care must be taken when comparing sounds of impulses. Often the big differences heard are in the HF range, not the LF range you are actually concerned with. It will be interesting to see if you find them more useful than the frequency/phase plots.

I sure would like to see before-and-after output traces for say speaker resonance and say, 80 Hz.
I may have saved a tone burst at resonance, will see what I find.

…what were your results using the second voice coil for feedback?
As I’ve posted several times over the years, pretty dismal. VC feedback from a DVC woofer can be used to flatten and extend response, but it is not capable of significant distortion reduction; often distortion is actually increased at 40Hz and above where BL related distortions dominate over the suspension distortions.
 
Since those measurements were calculated from their impulse response, this should be easy enough to post…assuming I didn’t pitch them. I’ll see what I can find this weekend. Care must be taken when comparing sounds of impulses. Often the big differences heard are in the HF range, not the LF range you are actually concerned with. It will be interesting to see if you find them more useful than the frequency/phase plots.

Thanks for all your help.

I trust actual pictures of sound waves much more than calculated curves and really have only a poor grasp of the many assumptions and circuits between the mic and screen output. (A lot of empirical researchers say that to a lot of theoretical scientists.)

About that "HF range", it's an old riddle and related to the perception of "fast" bass and how to interpret Dirac pulse and tone-burst images.

I think the answer may be this: if you can hear the "fast" bass (or see cleaner tone-burst images), then the actual sub-woofer pass-band of interest must also be more faithful to the input. Again, if the sound of an impulse goes from "clunk" to "click" when MFB is applied due mostly to better HF fidelity, then lower bass must be better too.

As I’ve posted several times over the years, pretty dismal. VC feedback from a DVC woofer can be used to flatten and extend response, but it is not capable of significant distortion reduction; often distortion is actually increased at 40Hz and above where BL related distortions dominate over the suspension distortions.

As has been debated in the past, there are several benefits of MFB or which struggling to get distortion lower is only one. I'd sure like to see if DVC feedback helps the speaker reaction to impulses, both the sound wave picture and the actual recorded sound.

If a person has a DVC woofer, then DVC feedback is (1) simple, (2) available to experiment for many readers (with a spare op amp) right away way, and (3) cheapest method.

B.
 
I appreciate that you may be more comfortable with one data format vs another…but it doesn’t make the other format wrong. It is all the same “actual” data, not theory. I managed to locate the parent impulse response data for those MFB experiments. Will get them imported and plotted in the next couple days.

The point of my cautionary comment about HF content is that once the response is bandwidth limited (upstream or downstream of the MFB loop) to subwoofing frequencies (ie < 80Hz) those HF differences you are hearing will no longer be conspicuous/relevant.

Although conceptually DVC MFB is simple, most DVC subwoofers on the market today are not good candidates for VC based MFB. Their high inductance and mutual coupling between the coils result in contamination of velocity signal from ~50Hz upward which is very difficult to compensate for. Your Brutus 15” woofer falls in this category.
Commercial motional feedback woofer available sort of

How a speaker with(or without) MFB reacts to impulses is just the frequency/phase response playing out in the time domain. Flattening and extending frequency response improves the impulse response. Improving the impulse response will flatten and extend frequency response. You can’t do one without the other. As gedlee mentioned to you, they are "intimately connected". Did you ever try the EQ measurement loop test I recommended to get a feel for how frequency response and transient response are tied to each other? Commercial motional feedback woofer available sort of
 
Last edited:
I appreciate that you may be more comfortable with one data format vs another…but it doesn’t make the other format wrong.

For a dissenting view, you might enjoy reading (or re-reading) Tufte famous book which is perhaps the most admired text in communications studies:

The Visual Display of Quantitative Information: Edward R. Tufte: 9781930824133: Amazon.com: Books

Let me ask you this. If I said the average height of people in Alaska is X inches, would that be right or wrong? What if I then told you the average height was Y for men and Z for women? You'd have to say the first "average" is absolutely right. By my view, it may be right in some sense, but quite wrong, misleading, and unproductive for anybody manufacturing coats. Or speaker stands, likewise.

...How a speaker with(or without) MFB reacts to impulses is just the frequency/phase response playing out in the time domain. Flattening and extending frequency response improves the impulse response. Improving the impulse response will flatten and extend frequency response. You can’t do one without the other....

"You can’t do one without the other" - that sort of pronouncement always calls for critical examination. For sure, it needs lots of boundary constraints to be the true in the least bit. How about a constraint such as "in the absence of active intervention", which is exactly the intervention we're talking about in this thread?

When I see an o'scope trace, I know exactly what I am looking at, given a basic description of the test set-up. But when I see a group delay created from a frequency sweep, I know I am looking at something that's been cooked 17 times over and none of us can be assured the software was working like we think we think it ought to be.

B.
 
Last edited:
… I managed to locate the parent impulse response data for those MFB experiments. Will get them imported and plotted in the next couple days.

OK, I located 4 different impulse measurements of interest:
(A) = Open loop response of woofer
(B) = Open loop response of woofer with analog 1kHz LP filter
(C) = Closed loop response of woofer with MFB using ACH-01
(D) = Open loop response of woofer with analog Linkwitz Transform EQ used to flatten and extend response to roughly match the closed loop.


Attached Plots:
  • Impulse Responses
  • Step Responses (calculated from Impulse)
  • Frequency Responses (calculated form Impulse)

Attached *.wav files:
  • Impulse of A
  • Impulse of B
  • Impulse of C
  • Impulse of D
  • Compare A vs C: A, A, pause, C, C, pause, A, A, pause, C, C
  • Compare B vs C: B, B, pause, C, C, pause, B, B, pause, C, C
  • Compare D vs C: D, D, pause, C, C, pause, D, D, pause, C, C

Enjoy :)
 

Attachments

  • MFB_Compare_Impulse.png
    MFB_Compare_Impulse.png
    78.7 KB · Views: 136
  • MFB_Compare_Step.png
    MFB_Compare_Step.png
    61.9 KB · Views: 136
  • MFB_Compare_Response.png
    MFB_Compare_Response.png
    38.2 KB · Views: 134
  • Impulse_wav_files.zip
    99.5 KB · Views: 25
Thanks for as elegant and compellingly logical a presentation as anyone could imagine.

I am struggling to parse the wonderful data. The attached impulse plots are hard to read on my gear. Any way to re-package the plots?

Being away from my home-base HiFi just now, using headphones seems to be more revealing of differences for me than poor available speakers; might be true for others too. I hope others will post their aural impressions.

My first gross impression is that LT-EQ and MFB produce similar FR and tone colour. That is a credit to both of them. LT-EQ works because Linkwitz analyzed right although as they say, "your results at home may vary". MFB works... which shows it works and possibly might not vary as much. LT-EQ is just cosmetic although simple to implement. MFB is a proper means of correction although complicated to implement. But I'm keen to closely eyeball how Bolserst impulse plots compare.

B.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.