Reply by October 30, 20042004-10-30
<allie@engr.wisc.edu> wrote:

>The person responsible for the signal enhancer you are refering to was Al >Clark. He now runs Danville Signal Processing which is where the recent >comp.dsp conference was held. dsp@danvillesignal.com I don't think Al >makes/sells these anymore. > >Yes an adaptive line enhancer is a good choice for adaptively separating the >correlated component of a signal from the uncorrelated. There are many >papers especially in the ieee journals on this technolgoy. Most Adaptive >filter text books cover this as well. One draw back on the ALE is the >bandwidth of the enhanced signal. The length of the adaptive filter >determines the bandwith of the ALE. I have generated what I call a recursive >ale to really narrow the bandwith of the ALE. I also noticed a paper at an >IEEE conference that refered to his recursive ALE as a feedback ALE or FALE. >I'm not kidding. If you search on FALE in the ieee conference papers it >shows up. An ALE, RALE or FALE will detect all periodic signals (within >reason). The drawback of the RALE of FALE is the reponse time to readapting >to changing periodic signals. In your case this doesn't seem to be a >problem. The frequency changes very slowly. You are in trouble if there are >periodic signals or signals with a longer correlation than your >decorellation delay that you want to keep. Feeding an adative filter with >the correlatred output of the ALE will work great. Maybe you already spoke >of this next method. I don't know I just tuned in. You can Bandpass filter >the incoming signal at 60 Hz(or what ever the fundamental frequency is). >Then determine the phase component by zero crossing detection (yuk) or using >a digital PLL or adative PLL to determine the real fundamental. This can be >used to generate a sin and cos of you fundamental. Then generate the >harmonics by multiplying the phase comp and generating more sine and cos >waves. This gives you separate signals for each frequency. These sin cos >pairs can be used to adapt a quadrature filter (only 1 tap per sin and 1 tap >per cos) using the original input as the error. It may not be more efficient >than the ALE method but it gives you more control over which tones are >removed because you can have a separate mu (covergence factor) for each. >Please note that when using a signal subtraction method like this a small >convergence factor should be used on the filter determining the amplitude >and phase of the signal to be subtracted. You can use a large convergence >factor but it will result in a non-linear filter which can actually change >the frequency of the reference signal if your reference doesn't exactly >match the error. Maybe that's not so bad. > >Mark Allie
Mark, Thanks for the additional information. Robert
Reply by October 29, 20042004-10-29
"John Monro" <johnmonro@delete.optusnet.com.au> wrote in message 
news:<417f0405$0$32541$afc38c87@news.optusnet.com.au>...

> r_obert@REMOVE_THIS.hotmail.com wrote:
>
> >John Monro <johnmonro@delete.optusnet.com.au> wrote:
> >
> >
> >
> >>r_obert@REMOVE_THIS.hotmail.com wrote:
> >>
> >>Robert,
> >>
> >>One technique is to use a delayed version of the signal itself as
> >>the
> >>reference input to the adaptive filter. This delay, often called a
> >>'decorrelation delay' ensures that the adaptive filter does not try to
> >>null out the changing part of the signal, but is able to null out the
> >>long-term ('coherent') part of the signal, i.e. the 60Hz component with
> >>its associated harmonics.
> >>
> >>
> >
> >That sounds promising.
> >
> >
> >
> >>The amount of decorrelation delay needs to be greater than the
> >>period of
> >>the lowest-frequency signal-component to be passed.
> >>
> >>
> >
> >Ah. So if the lowest frequency of the signal to-be-passed is 1 Hz,
> >the delay of the reference signal to the actual signal has to be at
> >least 1 second.
> >
> >
> >
> >>Hope this is of use,
> >>
> >>
> >
> >Yes, it has thanks. If you can think of a reference or paper to
> >describe this in more detail, that'd be great. But I'll just
> >initially approach it as a normal LMS or RLS adaptive filter
> >arrangement, but with the reference input simply delayed, per your
> >suggestion and the 'lowest-frequency signal-component' criteria.
> >
> >
> >
> >>regards,
> >>John
> >>
> >>
> >
> >Same,
> >
> >Robert
> >
> >www.gldsp.com
> >
> >
> Robert,
> The only reference I have on this is:
> Ifeachor & Jervis, Digital Signal Processing, A Practical
> Approach,Addison Wesley, 1993,
> p.544, chap. 9.2.2, "Other configurations of the adaptive
> filter." .
> The following page gives the topology of a number of different adaptive
> filter applications, including the one I described. Although chapter 9
> very good in its general coverage of adaptive filters, unfortunately
> there are no design notes given for these particular variations,
>
> Some time ago there appeared on the market a "signal enhancer" for Ham
> Radio use, which may be of interest to you.. I don't recall its
> actual name or maker, but it appeared to be based on two adaptive
> filters, in different configurations. One filter acted as discussed,
> to attenuate continuous tones in the received signal, exploiting the
> fact that this interference is more coherent than the speech. The
> part I particularly liked was that the other filter attenuated the
> wide-band interference by exploiting the fact that the 'hiss and
> crackle' interference is LESS coherent than the speech. Ifeachoe &
> Jervis show this configuration as well.
>
> Your intended aproach seems the right way to go.
>
> Good luck,
> regards,
> John
>
The person responsible for the signal enhancer you are refering to was Al Clark. He now runs Danville Signal Processing which is where the recent comp.dsp conference was held. dsp@danvillesignal.com I don't think Al makes/sells these anymore. Yes an adaptive line enhancer is a good choice for adaptively separating the correlated component of a signal from the uncorrelated. There are many papers especially in the ieee journals on this technolgoy. Most Adaptive filter text books cover this as well. One draw back on the ALE is the bandwidth of the enhanced signal. The length of the adaptive filter determines the bandwith of the ALE. I have generated what I call a recursive ale to really narrow the bandwith of the ALE. I also noticed a paper at an IEEE conference that refered to his recursive ALE as a feedback ALE or FALE. I'm not kidding. If you search on FALE in the ieee conference papers it shows up. An ALE, RALE or FALE will detect all periodic signals (within reason). The drawback of the RALE of FALE is the reponse time to readapting to changing periodic signals. In your case this doesn't seem to be a problem. The frequency changes very slowly. You are in trouble if there are periodic signals or signals with a longer correlation than your decorellation delay that you want to keep. Feeding an adative filter with the correlatred output of the ALE will work great. Maybe you already spoke of this next method. I don't know I just tuned in. You can Bandpass filter the incoming signal at 60 Hz(or what ever the fundamental frequency is). Then determine the phase component by zero crossing detection (yuk) or using a digital PLL or adative PLL to determine the real fundamental. This can be used to generate a sin and cos of you fundamental. Then generate the harmonics by multiplying the phase comp and generating more sine and cos waves. This gives you separate signals for each frequency. These sin cos pairs can be used to adapt a quadrature filter (only 1 tap per sin and 1 tap per cos) using the original input as the error. It may not be more efficient than the ALE method but it gives you more control over which tones are removed because you can have a separate mu (covergence factor) for each. Please note that when using a signal subtraction method like this a small convergence factor should be used on the filter determining the amplitude and phase of the signal to be subtracted. You can use a large convergence factor but it will result in a non-linear filter which can actually change the frequency of the reference signal if your reference doesn't exactly match the error. Maybe that's not so bad. Mark Allie John and Robert I appologize for sending this response to you directly and not using just the newsgroup.>
>
Reply by October 28, 20042004-10-28
John Monro <johnmonro@delete.optusnet.com.au> wrote:

>Robert, >The only reference I have on this is: > Ifeachor & Jervis, Digital Signal Processing, A Practical >Approach,Addison Wesley, 1993, > p.544, chap. 9.2.2, "Other configurations of the adaptive >filter." . >The following page gives the topology of a number of different adaptive >filter applications, including the one I described. Although chapter 9 >very good in its general coverage of adaptive filters, unfortunately >there are no design notes given for these particular variations,
Great, I'll see if I can't hunt that down.
> >Some time ago there appeared on the market a "signal enhancer" for Ham >Radio use, which may be of interest to you.. I don't recall its actual >name or maker, but it appeared to be based on two adaptive filters, in >different configurations. One filter acted as discussed, to attenuate >continuous tones in the received signal, exploiting the fact that this >interference is more coherent than the speech. The part I particularly >liked was that the other filter attenuated the wide-band interference >by exploiting the fact that the 'hiss and crackle' interference is LESS >coherent than the speech. Ifeachoe & Jervis show this configuration as >well.
Interesting. Is encouraging that there are proven practical applications. Sometimes there are great ideas proposed in papers or textbooks, but that run into snags in actual application. And then the question is sometimes not whether it can be made to work practically, but how much effort is required!
> >Your intended aproach seems the right way to go.
Great, that's just how I'll approach it.
> >Good luck, >regards,
Thanks, same. Robert www.gldsp.com
Reply by John Monro October 26, 20042004-10-26
r_obert@REMOVE_THIS.hotmail.com wrote:

>John Monro <johnmonro@delete.optusnet.com.au> wrote: > > > >>r_obert@REMOVE_THIS.hotmail.com wrote: >> >>Robert, >> >>One technique is to use a delayed version of the signal itself as the >>reference input to the adaptive filter. This delay, often called a >>'decorrelation delay' ensures that the adaptive filter does not try to >>null out the changing part of the signal, but is able to null out the >>long-term ('coherent') part of the signal, i.e. the 60Hz component with >>its associated harmonics. >> >> > >That sounds promising. > > > >>The amount of decorrelation delay needs to be greater than the period of >>the lowest-frequency signal-component to be passed. >> >> > >Ah. So if the lowest frequency of the signal to-be-passed is 1 Hz, >the delay of the reference signal to the actual signal has to be at >least 1 second. > > > >>Hope this is of use, >> >> > >Yes, it has thanks. If you can think of a reference or paper to >describe this in more detail, that'd be great. But I'll just >initially approach it as a normal LMS or RLS adaptive filter >arrangement, but with the reference input simply delayed, per your >suggestion and the 'lowest-frequency signal-component' criteria. > > > >>regards, >>John >> >> > >Same, > >Robert > >www.gldsp.com > >
Robert, The only reference I have on this is: Ifeachor & Jervis, Digital Signal Processing, A Practical Approach,Addison Wesley, 1993, p.544, chap. 9.2.2, "Other configurations of the adaptive filter." . The following page gives the topology of a number of different adaptive filter applications, including the one I described. Although chapter 9 very good in its general coverage of adaptive filters, unfortunately there are no design notes given for these particular variations, Some time ago there appeared on the market a "signal enhancer" for Ham Radio use, which may be of interest to you.. I don't recall its actual name or maker, but it appeared to be based on two adaptive filters, in different configurations. One filter acted as discussed, to attenuate continuous tones in the received signal, exploiting the fact that this interference is more coherent than the speech. The part I particularly liked was that the other filter attenuated the wide-band interference by exploiting the fact that the 'hiss and crackle' interference is LESS coherent than the speech. Ifeachoe & Jervis show this configuration as well. Your intended aproach seems the right way to go. Good luck, regards, John
Reply by October 26, 20042004-10-26
"Fred Marshall" <fmarshallx@remove_the_x.acm.org> wrote:

> >I've heard of that being used in a line enhancer (using the output of the >adaptive stage itself). If you use it as described above then you have to >make sure that the filter doesn't adapt too quickly or too slowly - as the >60Hz component has been described as changing as well. So, the signal of >interest would have to be moving faster than the 60Hz component that you >want to remove.
I'll take that at face value, but am not sure it's understood why.
>This would be a special case of interference and signal of interest that may >not always hold - even from time to time in the same system.
For power line noise, I believe the amplitude and phase could change quicker than the signal of interest ... but most likely not in frequency.
> >Maybe a silly implementation - but just for illustrative purposes - would be >a verrrry slowly changing >> non-changing adaptED filter: just set a >bandpass filter amplitude and phase to cancel the 60Hz interference. Then, >as long as the interference doesn't change, it will be cancelled by >subtraction. However, if the interference changes, the performance might >quickly degrade. And, the adapted filter would act on the signal.... if I >recall the context of this thread.
Interesting idea. Thanks, Robert
Reply by Fred Marshall October 26, 20042004-10-26
<r_obert@REMOVE_THIS.hotmail.com> wrote in message 
news:c0asn0l0rqr2vscq25eatglsvqbm7a7448@4ax.com...
> John Monro <johnmonro@delete.optusnet.com.au> wrote: > >>r_obert@REMOVE_THIS.hotmail.com wrote: >> >>Robert, >> >>One technique is to use a delayed version of the signal itself as the >>reference input to the adaptive filter. This delay, often called a >>'decorrelation delay' ensures that the adaptive filter does not try to >>null out the changing part of the signal, but is able to null out the >>long-term ('coherent') part of the signal, i.e. the 60Hz component with >>its associated harmonics.
I've heard of that being used in a line enhancer (using the output of the adaptive stage itself). If you use it as described above then you have to make sure that the filter doesn't adapt too quickly or too slowly - as the 60Hz component has been described as changing as well. So, the signal of interest would have to be moving faster than the 60Hz component that you want to remove. This would be a special case of interference and signal of interest that may not always hold - even from time to time in the same system. Maybe a silly implementation - but just for illustrative purposes - would be a verrrry slowly changing >> non-changing adaptED filter: just set a bandpass filter amplitude and phase to cancel the 60Hz interference. Then, as long as the interference doesn't change, it will be cancelled by subtraction. However, if the interference changes, the performance might quickly degrade. And, the adapted filter would act on the signal.... if I recall the context of this thread. Fred
Reply by October 26, 20042004-10-26
John Monro <johnmonro@delete.optusnet.com.au> wrote:

>r_obert@REMOVE_THIS.hotmail.com wrote: > >Robert, > >One technique is to use a delayed version of the signal itself as the >reference input to the adaptive filter. This delay, often called a >'decorrelation delay' ensures that the adaptive filter does not try to >null out the changing part of the signal, but is able to null out the >long-term ('coherent') part of the signal, i.e. the 60Hz component with >its associated harmonics.
That sounds promising.
> >The amount of decorrelation delay needs to be greater than the period of >the lowest-frequency signal-component to be passed.
Ah. So if the lowest frequency of the signal to-be-passed is 1 Hz, the delay of the reference signal to the actual signal has to be at least 1 second.
> > >Hope this is of use,
Yes, it has thanks. If you can think of a reference or paper to describe this in more detail, that'd be great. But I'll just initially approach it as a normal LMS or RLS adaptive filter arrangement, but with the reference input simply delayed, per your suggestion and the 'lowest-frequency signal-component' criteria.
>regards, >John
Same, Robert www.gldsp.com
Reply by John Monro October 25, 20042004-10-25
r_obert@REMOVE_THIS.hotmail.com wrote:

>"Fred Marshall" <fmarshallx@remove_the_x.acm.org> wrote: > > > >><r_obert@REMOVE_THIS.hotmail.com> wrote in message >> >> >>>The assumption here is that there is no way to make the noise >>>canceller adjust to frequency differences? >>> >>> >>I didn't elaborate much but I didn't mean to imply that. The assumption I >>made is that the "reasonable replica" might be from sensors that pick up a >>version of the noise - i.e. a linear combination of noises - such that >>amplitude and phase as a function of frequency may not be the same as the >>noise in the signal but a linear relationship exists. >> >> > >Yeah, that would be the standard arrangement. > > > >>So, if the noise changes frequency then so does the output of the canceller >>accordingly. I wasn't talking about synthesized noises but rather filtered >>versions of the real noise in real time. Otherwise, just as an example, >>let's assume that the noise is periodic. Then we could record one period of >>the noise and then use that as a reference to be subtracted (with suitable >>amplitude and phase changes). But, this wouldn't be a real time replica and >>not nearly as useful. >> >> > >Right. > >The reason I asked is because of facing a situation where I have the >signal + noise, but don't have access to anything correlated to the >noise, i.e. the power line, etc. But I know the noise is from power >line sources ... or at least the portion that I'm looking to remove >... and that it is sinusoidal ( with harmonics ), and in the range of >59-61 Hz. So the initial option appears to be fabricate a 60 Hz >sinusoidal signal as the reference input. But since it's uncorrelated >to the actual power line noise, it will not drift with it. And that >likely means that the noise canceller/adaptive filter will only remove >noise at 60 Hz. > >I guess the apparently unsolvable dilema is trying to remove power >line noise that drifts in frequency, without having an input reference >signal that is at least somewhat correlated to it. > >Regards, > >Robert > >www.gldsp.com > >
Robert, One technique is to use a delayed version of the signal itself as the reference input to the adaptive filter. This delay, often called a 'decorrelation delay' ensures that the adaptive filter does not try to null out the changing part of the signal, but is able to null out the long-term ('coherent') part of the signal, i.e. the 60Hz component with its associated harmonics. The amount of decorrelation delay needs to be greater than the period of the lowest-frequency signal-component to be passed. Hope this is of use, regards, John
Reply by October 25, 20042004-10-25
"Jon Harris" <goldentully@hotmail.com> wrote:

><r_obert@REMOVE_THIS.hotmail.com> wrote in message >news:8g5rn0t10ma69ml48do6fs72ltvjphiqir@4ax.com... >> "Fred Marshall" <fmarshallx@remove_the_x.acm.org> wrote: >> >> The reason I asked is because of facing a situation where I have the >> signal + noise, but don't have access to anything correlated to the >> noise, i.e. the power line, etc. But I know the noise is from power >> line sources ... or at least the portion that I'm looking to remove >> ... and that it is sinusoidal ( with harmonics ), and in the range of >> 59-61 Hz. So the initial option appears to be fabricate a 60 Hz >> sinusoidal signal as the reference input. But since it's uncorrelated >> to the actual power line noise, it will not drift with it. And that >> likely means that the noise canceller/adaptive filter will only remove >> noise at 60 Hz. >> >> I guess the apparently unsolvable dilema is trying to remove power >> line noise that drifts in frequency, without having an input reference >> signal that is at least somewhat correlated to it. > >There's always the good ol' notch filter. But usually there are harmonics >present as well, which complicates the matter...
Yeah, that's what's being used right now. But the notch is narrow, and so the attentuation suffers as the power line noise drifts ... plus the issue about missed harmonics, as mentioned. Regards, Robert www.gldsp.com
Reply by Jon Harris October 25, 20042004-10-25
<r_obert@REMOVE_THIS.hotmail.com> wrote in message
news:8g5rn0t10ma69ml48do6fs72ltvjphiqir@4ax.com...
> "Fred Marshall" <fmarshallx@remove_the_x.acm.org> wrote: > > The reason I asked is because of facing a situation where I have the > signal + noise, but don't have access to anything correlated to the > noise, i.e. the power line, etc. But I know the noise is from power > line sources ... or at least the portion that I'm looking to remove > ... and that it is sinusoidal ( with harmonics ), and in the range of > 59-61 Hz. So the initial option appears to be fabricate a 60 Hz > sinusoidal signal as the reference input. But since it's uncorrelated > to the actual power line noise, it will not drift with it. And that > likely means that the noise canceller/adaptive filter will only remove > noise at 60 Hz. > > I guess the apparently unsolvable dilema is trying to remove power > line noise that drifts in frequency, without having an input reference > signal that is at least somewhat correlated to it.
There's always the good ol' notch filter. But usually there are harmonics present as well, which complicates the matter...