FFT vs Terez frequency estimation method.
in [DSP]
|
Prev: TDM with PCM question
Next: From today's junk mail
From: Vladimir Vassilevsky on 21 Jun 2010 10:21 Dear Dmitry Teres, Just few questions: 1) Where can one get your algorithm for evaluation? Intentionally or unintentionally, your patent doesn't make any sense. 2) Is there any objective comparison of performance? 3) Can your claims be supported by a third party technical person, familiar with the art? Can you answer either question? fatalist wrote: > You can ignore all comments from VLV: he is a well-known abusive > comp.dsp troll, but what's even worse - he has no clue whatsoever > Can't even distinguish simple periodicity histogram from AMDF (!!!) > I bet he also calls autocorrelation AMDF and vice versa all the time Anything meaningfull to say? > I wonder who hires this dude ? My customers are very respectful companies. I do very well, and I don't hide my real name. Why are you hiding your name, Dmitry ? Are you ashamed of something? > (And this dude supposedly got his PhD from Fizteh - unbelievable, > simply unbelievable !!!) Yes, I got my degree from MiPh&T, and so did you, IIRC. So I can address the same question to you, my friend :)))))) VLV
From: tmtlib on 21 Jun 2010 12:02 i post code and signal later. Hope that my previous code may be usefull for someone at first time.
From: robert bristow-johnson on 22 Jun 2010 00:09 On Jun 21, 11:00 am, fatalist <simfid...(a)gmail.com> wrote: > On Jun 21, 8:53 am, Vladimir Vassilevsky <nos...(a)nowhere.com> wrote: > > > Hello Dmitry Teres's friend, how do you know it's his friend? it's more likely Dmitry himself posted this. > > > tmtlib wrote: > > > I have implemented simple modification of Dmitry Terez frequency estimation > > > method and get awesome results: > > > How much is awesome? something a little past objective. .... > > > > Incredible program. sounds objective. > > What it does is called AMDF, and this method is old as a world. well, AMDF with selective terms in the sum run through a non-linear function (the step function). > You DO realize that you are making a fool out of yourself, Vlad ? not to always come to Vlad's defense, but this is what i've been saying about Dmitry's alg since 2004 or something. it used to be that Google Groups could help me find these old posts, but it doesn't work so well anymore. in fact, today Google Groups works so shitty that i am not sure i can even post anymore. > > Here is your textbook AMDF definition: > > AMDF(k) = Sum |s(i)-s(i+k)| > > Here is periodicity histogram (in it's simplest one-dimensional form): > > hist(k)=sum H(r-|s(i)-s(i+k)|) (where H is Heaviside, or unit step, > function) > > Are you telling us that these two are the same ??? > > YES or NO ??? H(r-x) is a strictly decreasing function of x (r is a fixed parameter). it is nothing other than running the difference terms into a non-invertible non-linear function. since it's decreasing, instead of looking for a minimum as you would for AMDF, Dmitry is looking for a maximum after the sum. there are lot's of non-linear functions, why is H(r-x) the best one for this purpose they are not the same, and since H(|x|-r) is not the same as |x| nor | x|^2, we cannot say they are equivalent. but it *is* a subset of A[??]DF where [??] is a function of the magnitude of the difference. Vlad's right about this. i can say this because i have looked at Dmitry's paper and alg since 2002 or 2004 or whenever he first posted about this. r b-j
From: steveu on 22 Jun 2010 00:31 >On Jun 21, 11:00 am, fatalist <simfid...(a)gmail.com> wrote: >> On Jun 21, 8:53 am, Vladimir Vassilevsky <nos...(a)nowhere.com> wrote: >> >> > Hello Dmitry Teres's friend, > >how do you know it's his friend? it's more likely Dmitry himself >posted this. > >> >> > tmtlib wrote: >> > > I have implemented simple modification of Dmitry Terez frequency estimation >> > > method and get awesome results: >> >> > How much is awesome? > >something a little past objective. > >... >> >> > > Incredible program. > >sounds objective. > >> > What it does is called AMDF, and this method is old as a world. > >well, AMDF with selective terms in the sum run through a non-linear >function (the step function). > >> You DO realize that you are making a fool out of yourself, Vlad ? > >not to always come to Vlad's defense, but this is what i've been >saying about Dmitry's alg since 2004 or something. it used to be that >Google Groups could help me find these old posts, but it doesn't work >so well anymore. in fact, today Google Groups works so shitty that i >am not sure i can even post anymore. > >> >> Here is your textbook AMDF definition: >> >> AMDF(k) = Sum |s(i)-s(i+k)| >> >> Here is periodicity histogram (in it's simplest one-dimensional form): >> >> hist(k)=sum H(r-|s(i)-s(i+k)|) (where H is Heaviside, or unit step, >> function) >> >> Are you telling us that these two are the same ??? >> >> YES or NO ??? > >H(r-x) is a strictly decreasing function of x (r is a fixed >parameter). it is nothing other than running the difference terms >into a non-invertible non-linear function. since it's decreasing, >instead of looking for a minimum as you would for AMDF, Dmitry is >looking for a maximum after the sum. there are lot's of non-linear >functions, why is H(r-x) the best one for this purpose > >they are not the same, and since H(|x|-r) is not the same as |x| nor | >x|^2, we cannot say they are equivalent. but it *is* a subset of >A[??]DF where [??] is a function of the magnitude of the difference. Don't you means [??]MDF? Its using the magnitude difference, but its using a kind of twisted modal mean rather than an average. >Vlad's right about this. i can say this because i have looked at >Dmitry's paper and alg since 2002 or 2004 or whenever he first posted >about this. Steve
From: tmtlib on 22 Jun 2010 02:11
>On Jun 21, 11:00 am, fatalist <simfid...(a)gmail.com> wrote: >> On Jun 21, 8:53 am, Vladimir Vassilevsky <nos...(a)nowhere.com> wrote: >> >> > Hello Dmitry Teres's friend, > >how do you know it's his friend? it's more likely Dmitry himself >posted this. I am not Dmitry. My name is Georgy Moshkin. I've found Dmitry's method in google. > >> >> > tmtlib wrote: >> > > I have implemented simple modification of Dmitry Terez frequency estimation >> > > method and get awesome results: >> >> > How much is awesome? > >something a little past objective. > >... >> >> > > Incredible program. > >sounds objective. > >> > What it does is called AMDF, and this method is old as a world. > >well, AMDF with selective terms in the sum run through a non-linear >function (the step function). > >> You DO realize that you are making a fool out of yourself, Vlad ? > >not to always come to Vlad's defense, but this is what i've been >saying about Dmitry's alg since 2004 or something. it used to be that >Google Groups could help me find these old posts, but it doesn't work >so well anymore. in fact, today Google Groups works so shitty that i >am not sure i can even post anymore. > >> >> Here is your textbook AMDF definition: >> >> AMDF(k) = Sum |s(i)-s(i+k)| >> >> Here is periodicity histogram (in it's simplest one-dimensional form): >> >> hist(k)=sum H(r-|s(i)-s(i+k)|) (where H is Heaviside, or unit step, >> function) >> >> Are you telling us that these two are the same ??? >> >> YES or NO ??? > >H(r-x) is a strictly decreasing function of x (r is a fixed >parameter). it is nothing other than running the difference terms >into a non-invertible non-linear function. since it's decreasing, >instead of looking for a minimum as you would for AMDF, Dmitry is >looking for a maximum after the sum. there are lot's of non-linear >functions, why is H(r-x) the best one for this purpose > >they are not the same, and since H(|x|-r) is not the same as |x| nor | >x|^2, we cannot say they are equivalent. but it *is* a subset of >A[??]DF where [??] is a function of the magnitude of the difference. > >Vlad's right about this. i can say this because i have looked at >Dmitry's paper and alg since 2002 or 2004 or whenever he first posted >about this. > >r b-j > i will check admf. Also give windows program for analyzing mono wav files with Dmitry's method later. then maybe post implementation for adsp21xx dsp family, as an example for block floating point method fft replacement. I am sure that code will be much smaller than radix4. |