Hello, I am new to the group, thanks for having me. I am amazed at the breadth and depth of these DSP discussions, this newsgroup looks like a great resource. I'm glad to have found you guys. I am looking for a simple breakdown of the tradeoffs involved in a QMF implementation of a bandpass filter. As far as I know, a QMF implementation is overkill just to implement a BPF for use, for example, in an online biophysiological signal analysis application (for simple filtering, not for anything like sub-band coding, or pattern recognition and feature segmentation, etc). Sorry to paste some much, but the following has been brought to my attention and I'd like to assess the validity. I'm not necessarily saying these claims are false, I'm just wondering what the cost is for all these benefits (noise, phase distortion, complexity, performance, etc...). Thanks a lot, any comments are greatly appreciated. Marc Quadrature filters have the following advantages over conventional digital filters: - Readily adjustable center frequency: The center frequency of the filter is set by the local reference source, which is the frequency of a local sine wave. Therefore, as the filter center frequency is changed, there is no need to recompute the filter coefficients. The center-frequency is thus adjustable with arbitrary precision, and can be set to track any desired frequency. - Transient response independent of center frequency: The transient response is set by the low-pass filters alone. Therefore, a 3 Hz wide filter set at 4 Hz will have a transient response (ability to track rapid changes in the EEG) identical to that of a 3 Hz wide filter set at 15 Hz. As a result, all EEG components can be tracked with the same filter response characteristics, providing uniformity and consistency for EEG training. - No envelope detection delay: The quadrature filter computes, internally, the amplitudes of the inphase and out-of-phase components, as the outputs of the low-pass filters. These values can be combined directly, to produce the amplitude of the signal at any instant. Therefore, it is not necessary to perform an envelope detection operation on the filtered waveform. Envelope detection causes an additional delay, since it always lags the signal. - Perfectly symmetrical passbands, centered exactly at the center-frequency. - Guaranteed zero phase shift at the center of the passband.
QMF benefits/tradeoffs
Started by ●August 10, 2005
Reply by ●August 10, 20052005-08-10
Marc, This is a short (incomplete) response. 1) I think your are confusing the term quadrature mirror filtering (QMF) with what you are describing as quadrature filtering. 2) The quadrature filters you are describing can be standard LP digital filters placed after a complex mixer, and possibly before a second mixer. 3) There is delay in the filtering, and consequently in the envelope calculations. 4) If the mixing frequencies are fixed for each filter, bandpass filters at a desired center freq can often be designed by a simple mix of a set of LP filter coefficients. This results in less calculations that mixing the signal to baseband, filtering and remixing back to the original signal location (assuming multirate implementation not being used). Dirk Marc wrote:> Hello, > > I am new to the group, thanks for having me. I am amazed at the > breadth and depth of these DSP discussions, this newsgroup looks like a > great resource. I'm glad to have found you guys. > > I am looking for a simple breakdown of the tradeoffs involved in a QMF > implementation of a bandpass filter. As far as I know, a QMF > implementation is overkill just to implement a BPF for use, for > example, in an online biophysiological signal analysis application (for > simple filtering, not for anything like sub-band coding, or pattern > recognition and feature segmentation, etc). Sorry to paste some much, > but the following has been brought to my attention and I'd like to > assess the validity. I'm not necessarily saying these claims are > false, I'm just wondering what the cost is for all these benefits > (noise, phase distortion, complexity, performance, etc...). Thanks a > lot, any comments are greatly appreciated. > > Marc > > Quadrature filters have the following advantages over conventional > digital filters: > > - Readily adjustable center frequency: The center frequency of the > filter is set by the local reference source, which is the frequency of > a local sine wave. Therefore, as the filter center frequency is > changed, there is no need to recompute the filter coefficients. The > center-frequency is thus adjustable with arbitrary precision, and can > be set to track any desired frequency. > > - Transient response independent of center frequency: The transient > response is set by the low-pass filters alone. Therefore, a 3 Hz wide > filter set at 4 Hz will have a transient response (ability to track > rapid changes in the EEG) identical to that of a 3 Hz wide filter set > at 15 Hz. As a result, all EEG components can be tracked with the same > filter response characteristics, providing uniformity and consistency > for EEG training. > > - No envelope detection delay: The quadrature filter computes, > internally, the amplitudes of the inphase and out-of-phase components, > as the outputs of the low-pass filters. These values can be combined > directly, to produce the amplitude of the signal at any instant. > Therefore, it is not necessary to perform an envelope detection > operation on the filtered waveform. Envelope detection causes an > additional delay, since it always lags the signal. > > - Perfectly symmetrical passbands, centered exactly at the > center-frequency. > > - Guaranteed zero phase shift at the center of the passband.
Reply by ●August 11, 20052005-08-11
Hi Dirk, Thanks for the reply and for clearing up my confusion. You're saying the inherent delay of the filtering will translate to a delay in the envelop detection. Is the statement about zero phase lag at the center frequency accurate? Can anyone else please comment on the validity of the claims about quadrature filters in my original post and perhaps expand on the benefits/tradeoffs of using quadrature filters for BP filtering? Thanks a lot, Marc
Reply by ●August 11, 20052005-08-11
Marc wrote:> Hi Dirk, > > Thanks for the reply and for clearing up my confusion. You're saying > the inherent delay of the filtering will translate to a delay in the > envelop detection. Is the statement about zero phase lag at the > center frequency accurate?Phase delay relative to what? There is an inevitable _time_ delay.> Can anyone else please comment on the validity of the claims about > quadrature filters in my original post and perhaps expand on the > benefits/tradeoffs of using quadrature filters for BP filtering? > > Thanks a lot, > MarcQuadrature filters, or QMF filter banks? You seem to have the two muddled together. http://www.google.com/search?q=quadrature+mirror+filter Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●August 11, 20052005-08-11
Sorry about the confusion. I am in fact asking about the weaver filter shown in http://www.doug-smith.net/hifidesign1.htm - scroll down to fig 4. The claims in my original post were made about the benefits of such an implementation. Can anyone comment on their validity? Thanks, Marc
Reply by ●August 11, 20052005-08-11
Marc wrote:> Sorry about the confusion. I am in fact asking about the weaver filter > shown in http://www.doug-smith.net/hifidesign1.htm - scroll down to fig > 4. The claims in my original post were made about the benefits of such > an implementation. Can anyone comment on their validity?I just opened the page. I'll quote and comment as I read. "An SSB signal can [therefore] be considered a combination of amplitude modulation (AM) and phase modulation (PM)." Certainly, not directly. I suppose it could be considered the sum of several separately modulated signals, but why? "An SSB signal's envelope is identical to that of the baseband signal used to produce it." ??? A baseband signal doesn't have an envelope. "Were no dc present at baseband, modulator balance alone might be sufficient to knock down the carrier. But an asymmetrical baseband waveform may have an inherent dc component. Some voices or other sources may produce a baseband voltage that is positive more often than it is negative, or vice versa." In a linear circuit, asymmetry can't produce DC where there is none originally. If any frequency not in the input is present in the output, the circuit is not LTIV. "Capacitive coupling tends to mitigate that effect, but it cannot cope with short-term changes in dc content." Short-term changes in DC content are in fact low frequencies. If they weren't present in the signal, they are the result of distortion. I don't see any claims about or even discussion of filter properties. He may be onto something I don't understand. Whatever, I don't understand. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●August 11, 20052005-08-11
"Marc" <marc.saab@mail.mcgill.ca> wrote in message news:1123787917.139683.219100@g49g2000cwa.googlegroups.com...> Sorry about the confusion. I am in fact asking about the weaver filter > shown in http://www.doug-smith.net/hifidesign1.htm - scroll down to fig > 4. The claims in my original post were made about the benefits of such > an implementation. Can anyone comment on their validity? > > Thanks, > Marc >Hi Marc That reference was useful in understanding your questions better...with that as a background, let me see if I can address some of the claims...> Quadrature filters have the following advantages over conventional > digital filters:I've never heard of these being referred to as 'Quadrature filters'. The block diagram is essential a mixer followed by low pass filters. You can mix down any carrier frequency to baseband and low pass filter them...instead of band pass filtering them at the carrier freq - this is essentially what the block diagram is asserting.> - Readily adjustable center frequency: The center frequency of the > filter is set by the local reference source, which is the frequency of > a local sine wave. Therefore, as the filter center frequency is > changed, there is no need to recompute the filter coefficients. The > center-frequency is thus adjustable with arbitrary precision, and can > be set to track any desired frequency.Yes - by changing the mixer's LO, you get a tunable down-converter.> - Transient response independent of center frequency: The transient > response is set by the low-pass filters alone. Therefore, a 3 Hz wide > filter set at 4 Hz will have a transient response (ability to track > rapid changes in the EEG) identical to that of a 3 Hz wide filter set > at 15 Hz. As a result, all EEG components can be tracked with the same > filter response characteristics, providing uniformity and consistency > for EEG training.This is true since you only have 1 set of LP filters.> - No envelope detection delay: The quadrature filter computes, > internally, the amplitudes of the inphase and out-of-phase components, > as the outputs of the low-pass filters. These values can be combined > directly, to produce the amplitude of the signal at any instant. > Therefore, it is not necessary to perform an envelope detection > operation on the filtered waveform. Envelope detection causes an > additional delay, since it always lags the signal.I don't quite understand this - my guess is that this is referring to how one will have a delay if this was implemented using analog circuits vs *almost* no delay using this method. I guess I can buy that.> - Perfectly symmetrical passbands, centered exactly at the > center-frequency.The LP filters are essentially symmetrical around DC (although they can be made not to be). So you get this.> - Guaranteed zero phase shift at the center of the passband.Not sure about this. I think you also need to assess the impact of these 'claims'...do you really care? What are your design alternatives? You should compare these pros (and any cons) of this block diagram against other alternatives (I don't know what you are trying to do, so I don't know what others you can consider). Cheers Bhaskar
Reply by ●August 11, 20052005-08-11
Hi Bhaskar, Thanks for the detailed response. The issue is that I am trying to decide if I should emulate the design (due to all these wonderful benefits) or if I should simply criticize it (due to *some* detriments). Can you perhaps comment on some of the tradeoffs involved in such an implementation, such as increased computation time, complexity, ripple, estimation noise, or whatever else might be sacrificed for the aforementioned benefits. Since I don't know much about this sort of implementation, I decided to rely on you guys as my source of information. I was unable to find any other useful information or reference (most likely due to my lack of a clear definition of the implementation). Thanks. Marc
Reply by ●August 11, 20052005-08-11
Bhaskar Thiagarajan wrote:> "Marc" <marc.saab@mail.mcgill.ca> wrote in message > news:1123787917.139683.219100@g49g2000cwa.googlegroups.com... > >>Sorry about the confusion. I am in fact asking about the weaver filter >>shown in http://www.doug-smith.net/hifidesign1.htm - scroll down to fig >>4. The claims in my original post were made about the benefits of such >>an implementation. Can anyone comment on their validity? >> >>Thanks, >>Marc >> > > > Hi Marc > > That reference was useful in understanding your questions better...with that > as a background, let me see if I can address some of the claims... > > >>Quadrature filters have the following advantages over conventional >>digital filters: > > > I've never heard of these being referred to as 'Quadrature filters'. The > block diagram is essential a mixer followed by low pass filters. You can mix > down any carrier frequency to baseband and low pass filter them...instead of > band pass filtering them at the carrier freq - this is essentially what the > block diagram is asserting. > > >>- Readily adjustable center frequency: The center frequency of the >>filter is set by the local reference source, which is the frequency of >>a local sine wave. Therefore, as the filter center frequency is >>changed, there is no need to recompute the filter coefficients. The >>center-frequency is thus adjustable with arbitrary precision, and can >>be set to track any desired frequency. > > > Yes - by changing the mixer's LO, you get a tunable down-converter. > > >>- Transient response independent of center frequency: The transient >>response is set by the low-pass filters alone. Therefore, a 3 Hz wide >>filter set at 4 Hz will have a transient response (ability to track >>rapid changes in the EEG) identical to that of a 3 Hz wide filter set >>at 15 Hz. As a result, all EEG components can be tracked with the same >>filter response characteristics, providing uniformity and consistency >>for EEG training. > > > This is true since you only have 1 set of LP filters. > > >>- No envelope detection delay: The quadrature filter computes, >>internally, the amplitudes of the inphase and out-of-phase components, >>as the outputs of the low-pass filters. These values can be combined >>directly, to produce the amplitude of the signal at any instant. >>Therefore, it is not necessary to perform an envelope detection >>operation on the filtered waveform. Envelope detection causes an >>additional delay, since it always lags the signal. > > > I don't quite understand this - my guess is that this is referring to how > one will have a delay if this was implemented using analog circuits vs > *almost* no delay using this method. I guess I can buy that. > > >>- Perfectly symmetrical passbands, centered exactly at the >>center-frequency. > > > The LP filters are essentially symmetrical around DC (although they can be > made not to be). So you get this. > > >>- Guaranteed zero phase shift at the center of the passband. > > > Not sure about this. > > I think you also need to assess the impact of these 'claims'...do you really > care? What are your design alternatives? You should compare these pros (and > any cons) of this block diagram against other alternatives (I don't know > what you are trying to do, so I don't know what others you can consider).You don't get the magnitude of the carrier "directly" from I and Q, you get it from sqrt(I^2 + Q^2). There are faster ways to compute that than squaring, summing, and rooting, but it still takes several iterations. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●August 11, 20052005-08-11
Marc wrote:> Hi Bhaskar, > > Thanks for the detailed response. > > The issue is that I am trying to decide if I should emulate the design > (due to all these wonderful benefits) or if I should simply criticize > it (due to *some* detriments). Can you perhaps comment on some of the > tradeoffs involved in such an implementation, such as increased > computation time, complexity, ripple, estimation noise, or whatever > else might be sacrificed for the aforementioned benefits. Since I > don't know much about this sort of implementation, I decided to rely on > you guys as my source of information. I was unable to find any other > useful information or reference (most likely due to my lack of a clear > definition of the implementation).Do you want to make a ham transmitter? A wireless mic? Something else? How much does good low-frequency response matter to you? What do you call low? (Telephones go down to around 300 Hz; 150 is nicer. 20 is ridiculous for voice.) Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
