Hi all, I cannot really find some appropriate literature on de-emphasis done in digital domain. What I managed to understand from the web is that in analog domain, it is an first order RC low pass filter or H(s) = 1/(1+RCs). I have tried to turn this into digital representation by using bi-linear transform without prewarping and I think I do not get the right slope although I do get that cut off frequency (or -3dB power) is at 1/(2piRC) as expected. Is there a way to overcome this? Is there some literature that you are aware of, where I can read about digital implementation of de emphasis filter or preemphasis filter? I also have hard time finding out weather this preemphasis is even used nowadays. --------------------------------------- Posted through http://www.DSPRelated.com
De Ephasis Digital Filter
Started by ●December 17, 2015
Reply by ●December 17, 20152015-12-17
On Thu, 17 Dec 2015 07:55:04 -0600, b2508 wrote:> Hi all, > > I cannot really find some appropriate literature on de-emphasis done in > digital domain. > > What I managed to understand from the web is that in analog domain, it > is an first order RC low pass filter or H(s) = 1/(1+RCs). > > I have tried to turn this into digital representation by using bi-linear > transform without prewarping and I think I do not get the right slope > although I do get that cut off frequency (or -3dB power) is at 1/(2piRC) > as expected. > > Is there a way to overcome this? > Is there some literature that you are aware of, where I can read about > digital implementation of de emphasis filter or preemphasis filter? > I also have hard time finding out weather this preemphasis is even used > nowadays.Isn't there a paper by Geoffrey DeEmphasis, 3rd Baron of Emphasis, that goes over all of this? Seriously, could you post the z-domain transfer function you're generating? Unless you're working fairly close to the sampling rate your slope should be 20dB/decade. If you _are_ working fairly close to the sampling rate you can improve things somewhat by clever use of zeros in the transfer functions, but only to an extent. -- www.wescottdesign.com
Reply by ●December 18, 20152015-12-18
>On Thu, 17 Dec 2015 07:55:04 -0600, b2508 wrote: > >> Hi all, >> >> I cannot really find some appropriate literature on de-emphasis donein>> digital domain. >> >> What I managed to understand from the web is that in analog domain, it >> is an first order RC low pass filter or H(s) = 1/(1+RCs). >> >> I have tried to turn this into digital representation by usingbi-linear>> transform without prewarping and I think I do not get the right slope >> although I do get that cut off frequency (or -3dB power) is at1/(2piRC)>> as expected. >> >> Is there a way to overcome this? >> Is there some literature that you are aware of, where I can read about >> digital implementation of de emphasis filter or preemphasis filter? >> I also have hard time finding out weather this preemphasis is evenused>> nowadays. > >Isn't there a paper by Geoffrey DeEmphasis, 3rd Baron of Emphasis, that >goes over all of this? > >Seriously, could you post the z-domain transfer function you're >generating? Unless you're working fairly close to the sampling rate your>slope should be 20dB/decade. If you _are_ working fairly close to the >sampling rate you can improve things somewhat by clever use of zeros in >the transfer functions, but only to an extent. > >-- >www.wescottdesign.comI suppose this name is a joke and everyone knows actual name of the author, but I don't, so could you tell me please? :-) --------------------------------------- Posted through http://www.DSPRelated.com
Reply by ●December 18, 20152015-12-18
On 18.12.15 15:32, b2508 wrote:>> On Thu, 17 Dec 2015 07:55:04 -0600, b2508 wrote: >> >>> Hi all, >>> >>> I cannot really find some appropriate literature on de-emphasis done > in >>> digital domain. >>> >>> What I managed to understand from the web is that in analog domain, it >>> is an first order RC low pass filter or H(s) = 1/(1+RCs). >>> >>> I have tried to turn this into digital representation by using > bi-linear >>> transform without prewarping and I think I do not get the right slope >>> although I do get that cut off frequency (or -3dB power) is at > 1/(2piRC) >>> as expected. >>> >>> Is there a way to overcome this? >>> Is there some literature that you are aware of, where I can read about >>> digital implementation of de emphasis filter or preemphasis filter? >>> I also have hard time finding out weather this preemphasis is even > used >>> nowadays. >> >> Isn't there a paper by Geoffrey DeEmphasis, 3rd Baron of Emphasis, that >> goes over all of this? >> >> Seriously, could you post the z-domain transfer function you're >> generating? Unless you're working fairly close to the sampling rate your > >> slope should be 20dB/decade. If you _are_ working fairly close to the >> sampling rate you can improve things somewhat by clever use of zeros in >> the transfer functions, but only to an extent. >> >> -- >> www.wescottdesign.com > > I suppose this name is a joke and everyone knows actual name of the > author, but I don't, so could you tell me please? :-)Tim is attempting to tell that there is no better way of doing the simple filter, unless there is a god of the audio folks called by that name. If you want more discussion, please post the sampling frequency and the coefficients of your filter. -- -TV
Reply by ●December 18, 20152015-12-18
>>On Thu, 17 Dec 2015 07:55:04 -0600, b2508 wrote: >> >>> Hi all, >>> >>> I cannot really find some appropriate literature on de-emphasis done >in >>> digital domain. >>>[...snip...]>>> I also have hard time finding out weather this preemphasis is even >used >>> nowadays. >>>>Isn't there a paper by Geoffrey DeEmphasis, 3rd Baron of Emphasis, that>>goes over all of this? >>[...snip...]>> >>-- >>www.wescottdesign.com > >I suppose this name is a joke and everyone knows actual name of the >author, but I don't, so could you tell me please? :-) >--------------------------------------- >Posted through http://www.DSPRelated.comI'm sure the name is a joke, and I doubt a paper exists on it. You will, however, find tons of papers on digital filters. This post intrigued me so I did a little research. What I found is that pre-emphasis and de-emphasis, together called emphasis, is a technique for minimizing the impact of noise on a modulated signal introduced during transmission. The main principle being higher frequency components tend to have lower amplitudes so noise impacts them more. By amplifying the higher frequency components before transmission, then attenuating them on reception, causes any noise to be likewise attenuated leaving an overall hight SNR for those components. The second principle is the noise tends to be more powerful at the higher end of the spectrum. One huge advantage of going digital is that interpretation merely requires a threshold being reached. As long as any noise does not interfere significantly enough with the level to change the threshold being reached, the digital transmission will be 100% accurate, effectively noiseless. So I am very curious, other than as an academic exercise, why are you pursuing this topic? Ced --------------------------------------- Posted through http://www.DSPRelated.com
Reply by ●December 18, 20152015-12-18
"b2508" <108118@DSPRelated> writes:> Hi all, > > I cannot really find some appropriate literature on de-emphasis done in > digital domain. > > What I managed to understand from the web is that in analog domain, it is > an first order RC low pass filter or H(s) = 1/(1+RCs). > > I have tried to turn this into digital representation by using bi-linear > transform without prewarping and I think I do not get the right slope > although I do get that cut off frequency (or -3dB power) is at 1/(2piRC) > as expected. > > Is there a way to overcome this? > > Is there some literature that you are aware of, where I can read about > digital implementation of de emphasis filter or preemphasis filter? > I also have hard time finding out weather this preemphasis is even used > nowadays.Didn't I already refer you to the Frequency Domain Least Squares (FDLS) filter design technique (championed by Greg Berchin) when you ask a similar question some weeks ago? The answer hasn't changed. -- Randy Yates, DSP/Embedded Firmware Developer Digital Signal Labs http://www.digitalsignallabs.com
Reply by ●December 18, 20152015-12-18
>"b2508" <108118@DSPRelated> writes: > >> Hi all, >> >> I cannot really find some appropriate literature on de-emphasis donein>> digital domain. >> >> What I managed to understand from the web is that in analog domain, itis>> an first order RC low pass filter or H(s) = 1/(1+RCs). >> >> I have tried to turn this into digital representation by usingbi-linear>> transform without prewarping and I think I do not get the right slope >> although I do get that cut off frequency (or -3dB power) is at1/(2piRC)>> as expected. >> >> Is there a way to overcome this? >> >> Is there some literature that you are aware of, where I can read about >> digital implementation of de emphasis filter or preemphasis filter? >> I also have hard time finding out weather this preemphasis is evenused>> nowadays. > >Didn't I already refer you to the Frequency Domain Least Squares (FDLS) >filter design technique (championed by Greg Berchin) when you ask a >similar question some weeks ago? > >The answer hasn't changed. >-- >Randy Yates, DSP/Embedded Firmware Developer >Digital Signal Labs >http://www.digitalsignallabs.comYes youy have, but I have not found adequate literature. --------------------------------------- Posted through http://www.DSPRelated.com
Reply by ●December 18, 20152015-12-18
"b2508" <108118@DSPRelated> writes:>>"b2508" <108118@DSPRelated> writes: >> >>> Hi all, >>> >>> I cannot really find some appropriate literature on de-emphasis done > in >>> digital domain. >>> >>> What I managed to understand from the web is that in analog domain, it > is >>> an first order RC low pass filter or H(s) = 1/(1+RCs). >>> >>> I have tried to turn this into digital representation by using > bi-linear >>> transform without prewarping and I think I do not get the right slope >>> although I do get that cut off frequency (or -3dB power) is at > 1/(2piRC) >>> as expected. >>> >>> Is there a way to overcome this? >>> >>> Is there some literature that you are aware of, where I can read about >>> digital implementation of de emphasis filter or preemphasis filter? >>> I also have hard time finding out weather this preemphasis is even > used >>> nowadays. >> >>Didn't I already refer you to the Frequency Domain Least Squares (FDLS) >>filter design technique (championed by Greg Berchin) when you ask a >>similar question some weeks ago? >> >>The answer hasn't changed. >>-- >>Randy Yates, DSP/Embedded Firmware Developer >>Digital Signal Labs >>http://www.digitalsignallabs.com > > Yes youy have, but I have not found adequate literature.Email me privately and I may be able to help you out there. I implemented the entire BTSC decoder (MTS stereo decoder for television), which originally utilized analog filters, using this technique and the results were superb. -- Randy Yates, DSP/Embedded Firmware Developer Digital Signal Labs http://www.digitalsignallabs.com
Reply by ●December 18, 20152015-12-18
>On Thu, 17 Dec 2015 07:55:04 -0600, b2508 wrote: > >> Hi all, >> >> I cannot really find some appropriate literature on de-emphasis donein>> digital domain. >> >> What I managed to understand from the web is that in analog domain, it >> is an first order RC low pass filter or H(s) = 1/(1+RCs). >> >> I have tried to turn this into digital representation by usingbi-linear>> transform without prewarping and I think I do not get the right slope >> although I do get that cut off frequency (or -3dB power) is at1/(2piRC)>> as expected. >> >> Is there a way to overcome this? >> Is there some literature that you are aware of, where I can read about >> digital implementation of de emphasis filter or preemphasis filter? >> I also have hard time finding out weather this preemphasis is evenused>> nowadays. > >Isn't there a paper by Geoffrey DeEmphasis, 3rd Baron of Emphasis, that >goes over all of this? > >Seriously, could you post the z-domain transfer function you're >generating? Unless you're working fairly close to the sampling rate your>slope should be 20dB/decade. If you _are_ working fairly close to the >sampling rate you can improve things somewhat by clever use of zeros in >the transfer functions, but only to an extent. > >-- >www.wescottdesign.comhttps://www.fortnox.se/ Hi, I am sorry, I am not native english speaker or dsp professional so I do not understand your sarcasm and sometimes your answers either. I am trying to implement H(s) = 1/(1+tau*s) where tau = 50us in digital domain. My f_sampling is 100kHz. What I get in digital domain after bilinear transform is H(z) = (1+z^-1)/((1+2*tau/T)+(1-2*tau/T)*z^-1)) = (1+z^-1)/(11-9*z^-1) So coefficients are: b0=1, b1=1, a0= 11, a1=-9 I then tried to filter dirac of certain length with this and filter and plot FFT in log scale. I do not know how strict this 6db/octave rule should be but it seems to me that i dont get correct slope of the filter. --------------------------------------- Posted through http://www.DSPRelated.com
Reply by ●December 18, 20152015-12-18
On Fri, 18 Dec 2015 09:22:54 -0600, b2508 wrote:>>On Thu, 17 Dec 2015 07:55:04 -0600, b2508 wrote: >> >>> Hi all, >>> >>> I cannot really find some appropriate literature on de-emphasis done > in >>> digital domain. >>> >>> What I managed to understand from the web is that in analog domain, it >>> is an first order RC low pass filter or H(s) = 1/(1+RCs). >>> >>> I have tried to turn this into digital representation by using > bi-linear >>> transform without prewarping and I think I do not get the right slope >>> although I do get that cut off frequency (or -3dB power) is at > 1/(2piRC) >>> as expected. >>> >>> Is there a way to overcome this? >>> Is there some literature that you are aware of, where I can read about >>> digital implementation of de emphasis filter or preemphasis filter? >>> I also have hard time finding out weather this preemphasis is even > used >>> nowadays. >> >>Isn't there a paper by Geoffrey DeEmphasis, 3rd Baron of Emphasis, that >>goes over all of this? >> >>Seriously, could you post the z-domain transfer function you're >>generating? Unless you're working fairly close to the sampling rate >>your > >>slope should be 20dB/decade. If you _are_ working fairly close to the >>sampling rate you can improve things somewhat by clever use of zeros in >>the transfer functions, but only to an extent. >> >>-- >>www.wescottdesign.com > https://www.fortnox.se/ > Hi, > > I am sorry, I am not native english speaker or dsp professional so I do > not understand your sarcasm and sometimes your answers either.I'm sorry to inflict the humor on you, then. You wanted to title your post "De-emphasis Digital Filter". As written, it looks like a proper name of noble French extraction: DeEmphasis (or possibly d'Emphasis). So I made up an English baron (Geoffrey DeEmphasis) whose ancestors were no doubt thugs in the train of William the Conquerer (the more I type the more I realize this is a seriously English-culture in-joke), but who later became English gentry, probably under King George the 1st (see, even more in-joke).> I am trying to implement H(s) = 1/(1+tau*s) where tau = 50us in digital > domain. > My f_sampling is 100kHz. > > What I get in digital domain after bilinear transform is > > H(z) = (1+z^-1)/((1+2*tau/T)+(1-2*tau/T)*z^-1)) = (1+z^-1)/(11-9*z^-1) > > So coefficients are: b0=1, b1=1, a0= 11, a1=-9 I then tried to filter > dirac of certain length with this and filter and plot FFT in log scale. > I do not know how strict this 6db/octave rule should be but it seems to > me that i dont get correct slope of the filter.Your bilinear transform looks good. You can plot the frequency response much more directly by substituting w = e^(j * 2 * pi * T * f) then calculating H(w) at that frequency. Here f is your frequency of interest. Your corner frequency is going to be greater than 1/10 of your sampling rate, which points to the frequency vs. amplitude plot being more distorted from what you'd expect in the Laplace domain. I think your fundamental problem is coming from the fact that you're operating so close to the sampling rate. This gets back to what I said before: you _may_ be able to improve things by jiggering the zero around (the 1 + 1/z term in the numerator -- try <something else> + 1/z and see if you're happier). Beyond that, we're getting into the realm where I throw up my hands and start explaining that I know how to make motors do what I want, but I'm not really an audio processing guy. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com
