Hi All,

first of all, a big thank you for this forum as it is extremely useful expecially for a newbie like me.

I am in the process of creating a patch that could generate an inverse filter for an acoustic test signal (Log Swept sine) and then perform a deconvolution of the recording of the signal in the space.

What I am trying to do i recreating the swept sine from its mathematical formula which give me the exact inverse filter.

the formula is quite complex:

t)= sin(((w1*T)/(ln(w2/w1))*(e^(((T-t)/T)*log(w2/w1))-1))*(w2/w1)^(-t/T)

where t= time variable
T=signal lenght in s
w2= end frequency(normally 20 kHz)
w1= start frequency (normally 20 Hz)

Attached there is an initial patch that doesn't work.

Convolution engine might follow but for the moment I am more interested in generating the inverse filter.

Anyone has had experience with this?

I will definitely share the patch when finished

Thank you

http://www.pdpatchrepo.info/hurleur/Swept_sine_inverse.pd

Hello Katja,

this is great!!!

thank you very much.

there are still a couple of things to work out by the way:

1. at the end of the sweep there an unusual oscillation (probably my speakers but not sure) and I remember I have read somewhere that in generating the sine sweep there was the need to filter the highest frequency; i will try to find the paper and post it here (hopefully).

2. I tried to change the parameters in your code in order to have a longer sweep (2.1^7 = 6 sec at 44100) but the signal doesn't go till the end but stops earlier; so for sure it is my very newbie atempt to change your code but do you think it would be possible to create longer sweeps in PD?
   It is the main issue in my attempt of coding as normally when measuring acoustic environments you need very long sweep to measure everything you need (about 12 sec sometimes)

THis is great work anyway!!!

I will try to have a more in depth look these days to see if i can help you out with more specific questions.

Thank you

Good Night

I have modified the patch in such a way that start and stop frequency can be set at will, and table length can be 2^17, 2^18 or 2^19. It should be a power of two, to fit exactly in an fft frame.

With longer table length and lower stop frequency, it is easier to hear what actually happens. At the end, the signal stops suddenly of course, and the discontinuity produces a kind of 'blip' effect. Maybe we should just create a very short (64 sample) smooth fade-out there, instead of filtering.

It is now easy to produce log sweeps according to our specifications of frequency and length. But the longer sweeps cannot be handled in frequency domain (for inversion) within Pd. 2^17 really is the largest fft size here.

So, let's see if we can get the whole deconvolution thing done with that size. And if so, we can produce a longer sweep plus inverse in another software. After all you need to compute them only once. The rest of the work is for [partconv~] which handles fast convolution in partitions.

Katja

http://www.pdpatchrepo.info/hurleur/logsweep2.pd.zip

Cripes, this is really coming on! - looking forward to having a dig through when I've got free time.

Hello Katja and lead,

Katja - the new version of the sweep code is great!!!

I have taken some time at work now to calculate the exact exponential of 2 to have 3s, 6s, 12s and 24s sweep and this is included in the attached zip file (I have altered your patch, Katja).

regarding the generated sweep there are still two things to note:

1. at the end you have this 'bleep effect' and I agree with Katja that we can create a very short (64 sample) smooth fade-out

2. by the end of the sweep you get a frequency modulation (oscillation mainly) that is a bit strange; this can be due to the fact that I am listening on headphones the signal and I can have an acoustic "allucination" (typical when listening a mono test signal in a stereo set) or due to something in the code...
   comparing it to the signal at this link it doesn't appear to me

http://www.4horsemen.net/binkster/tracks/track06.zip

So probably we need to have a think about the audio output of the patch.

Sweep signal are mono signal and they are commonly reproduced through mono or multi channel mono systems (at least for room acoustics measurements).

Then regarding the convolution, It would be nice to create a self-working measurement environment in PD (so provide convolution as well) but let's discuss what we can do.

I will investigate a bit more the theory behind the deconvolution of the recorded signal with the inverse of the "pure" text signal and see if we can run the convolution in time domain only instead of occupying the frequency domain.

This is really exciting for me

Thank you very much for your effort in this

Bassik

I am having issues with the forum....

Cannot attach the modified patch, I am afraid.
I would try later or if you would like I can email it.

Ciao

Bassik

Hey Bassik,

I have now better listened to the long sweeps, and found the cause of the modulations. It is, as so often, a 32 bit precision issue. Here are 10 consecutive sample-indexes, as expressed in seconds:

print: 11.3397
print: 11.3397
print: 11.3397
print: 11.3398
print: 11.3398
print: 11.3398
print: 11.3398
print: 11.3398
print: 11.3399
print: 11.3399
print: 11.3399
print: 11.3399

With 64 bit floats and 13 decimal places this would be no problem:

11.33786848073
11.33789115646 etc

Bummer. I don't think we can get around this, with Pd.

Katja

Hello Katja,

thanks for the check.

We might be able to live with this issue, all depends on the (de)convolution process result.

If it eliminates the modulations as well we might be fine, if not we would need to find another way of building the sinesweeps.

Would you be able to do the same patch for the inverse signal, please?

I would try to have a go as well whenever I would get a second.

I will never stop to thank you for your interest.

Bassik

Dear bassik:
I read your post and a question arose in me when I read this:

"Sweep signal are mono signal and they are commonly reproduced through mono or multi channel mono systems (at least for room acoustics measurements)."

Question: Is it possible that some 'comb filter'-like effect appears when reproducing a mono signal through multiple loudspeakers for acoustic measurements? To my humble knowledge, early delays from each source might get to the microphone at different times for each frequency due to source-wall-mic distances, thus creating the 'comb-filter' effect.

Besides, your team work with @katjav and @lead is great. I wish you all success.

sumidero

I saw, many years ago, a PC-XP based ISA card for acoustic measurement of loudspeakers that instead of using a sine sweep it sent short trains of frequency fixed sines, opening the record just when the wave train was passing through the microphone.
The set-up was something like this:

PC --- [< c=~~~ (mic)
Loudspeaker |--0.3m--|

The record was set to start just at t0 = 0.3m / (sound speed in m/s) and ended just before the sine train ended to avoid the recording of other noises. The frequencies were stepped up instead of making a sine. Then the inverse calculus was done to measure the loudspeaker dB response.
I did this some years after that to make some sound treatment to a very small room (with many many normal modes!) which I use for audio creation. It sort of worked decently.
I didn't know about Pd at that time, so I created a long train of ((sine(n*w0*t)-silence-)xn freq bins) with Octave, used a DAW for the playing-recording and later recovered the dB response function with Octave again.

Could it be a choice/alternative for what you want to do?

Hugs to everyone...

sumidero.

@sumidero said:

Could it be a choice/alternative for what you want to do?

I have not heard of this technique yet. Do you know if it is mathematically described somewhere? How many frequencies, are those sine shots windowed, how is the inverse calculated, etc?

At the moment we even have problems with a technique which is described in detail, and one of the problems is the lack of precision in 32 bit floating point format, hindering us in making long sweeps.

I have now done a different implementation of log sweep calculation, using Farina's formula instead of the Matlab code. With Farina's formula, the imprecision pain is spread over the array instead of concentrating in the high indexes. Still not perfect but better than before, see patch.

We're still at step one! Unbelievable how time-consuming this is. In the 'Hamburg' article from Bassik's links is a method for amplitude inversion in time domain. I do not understand it rightaway, have to study a bit more on it.

http://www.pdpatchrepo.info/hurleur/logsweep4.pd.zip

Hello All,

this is becoming even more exciting!!

@Sumidero:

"Question: Is it possible that some 'comb filter'-like effect appears when reproducing a mono signal through multiple loudspeakers for acoustic measurements?"

Assuming you are in an Airport departure concourse.
You have a big multichannel PA system that should provide voice messages to all the people in the concourse.
You will have a guy speaking to a mic and says something like "Flight xxx will depart from gate yyy".
This is a mono signal reproduced by a multi-channel system.

Comb filtering are likely to happen but this is what you would like to measure to understand if the system would work and what are the downturns on having such a big PA.

Sumidero wrote: "I saw, many years ago, a PC-XP based ISA card for acoustic measurement of loudspeakers that instead of using a sine sweep it sent short trains of frequency fixed sines, opening the record just when the wave train was passing through the microphone."

is this system you are referring to?

http://www.mlssa.com/

IF yes, MLS (Maximum Lenght Signal) measurement system have been the industry standard until the sweep sine method comes out (expecially the log swept sine).

According to my knowledge, It is a very good way of measuring IRs but it has some problems in handling the non-linearities of the reproduction and recording chain.

@Katja

I will have a look to your patch and post some comments later, thank you very much.

I had a read through the 'Hamburg' article and he actually give us the exact function for the inverse swept sine which includes the amplitude inversion (or amplitude reduction, depends how you call it).
The way it is written is a bit confusing, I know, but after a bit of thinking you can see it like that:

assuming your funstion has a development in time (t) that goes between 0 and L-1.
the inverse function including time inversal and amplitude scaling is:

x^-1(t)= L-1-t)*(w2/w1)^(-t/L-1)

so if you want to write as written on physics books where your time variable t goes between 0 and T, it is like that:

x^-1(t)= T-t)*(w2/w1)^(-t/T) where 0<t<T

the function T-t) is basically the expression of the test signal (that you find previously in the paper) where instead of n (or t) you got L-1-n (or T-t).

this T-t) is expressed like that:

T-t)= sin(((w1-T)/log(w2/w1))*((e^((T-t)/T)*log(w2/w1)))-1)

I hope it is much more understandable now; I tried to use the same way of writing it in excel.
If you look at the xls file in the zip file i posted previously the formula is written properly and there is a graph showing the result.

This is very time consuming I know but there is no rush in finishing it as it could be an important tool in the free software community, I reckon.

Have a nice weekend guys

Bassik

@bassik said:

the function T-t) is basically the expression of the test signal

That's what I missed last night, thanks.

So the amplitude scaling is simply (w2/w1)^(-n/(L-1)) and this curve is multiplied with the time-inversed sweep. This is what I have implemented in Pd today, and it looks like the result you have in Excel. It also sounds like a sweep but of course the low frequencies are so soft you can't hear them. See patch.

However there is also this scaling constant C which must be applied to compensate power loss. This is the part that I can't get done. Note that frequencies w1 and w2 are expressed as dimensionless frequencies within range 0 till pi in this text. With equation (8) for the sweep I get good result, but equation (10) for the scaling constant gives excessively large results, due to arraylength L in the nominator. Can you figure this out?

If we have the scaling factor right we can start building the deconvolution engine. Then, if tests are promising, we should find a way to generate better sweeps. This can be done in a specialized external [logsweep] for example, using doubles instead of floats and writing results into named arrays. But this is only worth the effort if we don't find more barriers on our way.

Katja

http://www.pdpatchrepo.info/hurleur/logsweep&inv.pd.zip

@katjav said:

@sumidero said:

Could it be a choice/alternative for what you want to do?

I have not heard of this technique yet. Do you know if it is mathematically described somewhere? How many frequencies, are those sine shots windowed, how is the inverse calculated, etc?

As far as I can remember, there were no phase calculations involved in my calculi, just relative amplitudes for each frequency bin, in dB. I had set the loudspeakers outdoors, side by side and heading upwards to the sky, placed the mic a meter above them and got the 'reference curve' trusting the PA system response (a pair of close field reference monitors), then did the same inside the room. Response over difference in dB and hands on craftmanship to make bass traps. I know nowadays that I maybe should have done something better, with better equipment, I know it was sort of rough, but at that time I just wanted a room with a flatter response, asap.

Nowadays my job has gotten a little hectic, sorry for the delay in giving you feedback.
Best regards,

sumidero

@bassik said:

---

Sumidero wrote: "I saw, many years ago, a PC-XP based ISA card for acoustic measurement of loudspeakers that instead of using a sine sweep it sent short trains of frequency fixed sines, opening the record just when the wave train was passing through the microphone."

is this system you are referring to?

http://www.mlssa.com/

IF yes, MLS (Maximum Lenght Signal) measurement system have been the industry standard until the sweep sine method comes out (expecially the log swept sine).

According to my knowledge, It is a very good way of measuring IRs but it has some problems in handling the non-linearities of the reproduction and recording chain.

Yes, I think it was it. It was shown on an old CGA monitor but it looks very similar.

I think that one of the main drawbacks of that method might be the influence of the transients in each sine train. If you don't cut them out they would introduce artifacts in the spectra. At the time I did my home measurements, I checked these transients throughout the frequency spectrum and made an algorithm to slice them out before doing dB calculi. I wander where that Octave code has gone to...

I wish you all a good week.

sumidero

Made some good progress on the topic at last.

Attached patch has a log chirp generator for max 20 seconds, and computes the inverse. As a test, the chirp is convolved with it's own inverse, using [partconv~]. The result should approximate a single pulse of value 1 at time L = chirplength.

With stop frequency set at sr/2, that pulse is 0.997393 and surrounding values are in the order of 1/1000. With lower stop frequency, results quickly drift away from the ideal case. The imprecision artefacts as we perceived them, seem to do relatively little harm.

I think that the mathematical routines, though not perfect, are far more precise than physical test equipment (speakers, microphones) will ever be.

So it is now a matter of including record / load / save routines, to collect some real-life responses.

Katja

http://www.pdpatchrepo.info/hurleur/logsweep&inv-test.pd

Hello Katja,

that is great news!!!

I was looking into the scaling factor today but get stopped by work stuff and I am planning to have a look tonight.

I will also test the patch as well.

If you want I can do a quick real world test here at work as I have all the equipment.

In my first attachment there was a sort of implementation of reproduce and recording of the signal (if I remember well), would this be of help??

Bassik

Also forgot to mention that looking around for another reason I found this software made in Java:

http://www.hometheatershack.com/roomeq/

It does IR measurements as well.

Cheers

Bassik

Included some recording-, save- and load-routines to the log sweep deconvolution, and done a first real world test in my room. Dreadful standing waves oh boy.

Status quo patch attached.

Some weighted amplitude normalisation should be done on the sweep response. Now I have done an easy peak normalisation, but this will always lead to IR's having too low power.

Anyway, it is already possible to see the effect of the log sweep method (by the way this works with a linear sweep as well): non-linear distortion and pollution is shifted left of time zero and thus separated from the impulse response proper. Use vertical zoom to see this.

It would be useful to include an elementary wave editor for trimming. I have now done this with Audacity. Hardoff has done a wave editor in Pd, we could use a subset of it and include vertical zoom.

Katja

http://www.pdpatchrepo.info/hurleur/logsweepdeconv-test.pd.zip

@bassik said:

I found this software made in Java:

http://www.hometheatershack.com/roomeq/

It's great and it does all you need plus more.

Our Pd patch may not grow to such a professional tool, if only because it would require years of developement. But from an educational perspective a Pd implementation may be more interesting. I remember very well how IR measurement was demonstrated by our dsp teacher in Sonology class. We went through the mathematics, and I was excited about how you can expand and compress time dimension with help of a sweep.

With our Pd patch, the whole process can be visualised. At any point in it, you can test how (im)precise your method and result is. You can then decide whether it is good enough for your purpose, or should be improved. You can read Ben Saylor's [partconv~] code to see how fast convolution is done in practice. The benefits of open source. I see good reason to continue our log sweep deconv project.

Katja

Hello Katjav,

The new patch is great, it works well and already does more than I was expected to reach in such a small amount of time.

I would like to do some tests and comparison with a commercial software with the same functionality but I have just realised that the laptop with the software installed is not available until monday; nevermind.

I have two questions on the patch:

1. there is a strange behaviour in PD, if i turn on dsp the patch doesn't allow me to move the horizontal slides or better it doesn't visualize the movements while with dsp off there is no problem. It is a bit of a problem as you don't see the computation progress slider moving.
   Do you know why?
   No error messages and the last PD installed.

2. would you please expand a bit more about weighted amplitude normalisation?
   in the case of IRs normalising the main peak of the file will make all the IR information readable, or probably I am missing the point here.

so regarding the to-do list:

• Multichannel: it is quite important it could be multichannel input (i.e. ambisonic mic or stereo pairs) while it is much less important as multichannel output (1 signal would be common to all speakers you are testing, no need for different signal).

• fft analysis of IR: great idea but would be great to derive also additional acoustic parameters, I will provide maths for all the parameters we need to implement (ISO 3382 part 1 and part2, sorry I need to send the papers privately as I would get in trouble if I share on the web).

Also I would like to add in the future, a-format to b-format encoding (http://pcfarina.eng.unipr.it/Public/B-format/A2B-conversion/A2B.htm).
This is why 4 capsules tetrahedral mics can be used to analyse 3D soundscapes including directional analysis (along 3 axes) and understand issues in room such as room modes directions, flutter echoes and first reflections direction among others.

I am glad you have done a clean-up of the scaling constant calculation as in the previous patch it was a bit confusing to me.

I will have a more in depth play around tomorrow or friday...i am also in the middle of moving house...:(

Thank you

Bassik