Hello People I have one question to ask. In real time environment, we face multipath, doppler, noise etc. Now when we perform simulations, we plot BER vs SNR plot. In that the SNR is pure SNR (pure in the sense SNR=Eb/No, where we just take noise into account). How can we estimate this pure SNR, when we have got real time received signal? Thanking you Chintan
Estimate SNR
Started by ●June 2, 2008
Reply by ●June 2, 20082008-06-02
>Hello People > >I have one question to ask. > >In real time environment, we face multipath, doppler, noise etc. > >Now when we perform simulations, we plot BER vs SNR plot. In that theSNR>is pure SNR (pure in the sense SNR=Eb/No, where we just take noise into >account). > >How can we estimate this pure SNR, when we have got real time received >signal? > >Thanking you > >Chintan >Hi Chintan, one usually knows the power of the transmited signal as well as the power of the additive noise at the receiver. Estimation can be done once a data model is formulated relating the transmited and received signals. The data model can be simple to complicated, depending on how many phenomena acounts for. For example, in some cases an appropriate model for the channel is an FIR filter h[n]. If the transmited signal is s[n] and the additive noise at the receiver is w[n], then the data model is y[n] = conv(h[n],s[n]) + w[n], where y[n] is the received signal. Manolis
Reply by ●June 2, 20082008-06-02
>>Hello People >> >>I have one question to ask. >> >>In real time environment, we face multipath, doppler, noise etc. >> >>Now when we perform simulations, we plot BER vs SNR plot. In that the >SNR >>is pure SNR (pure in the sense SNR=Eb/No, where we just take noise into >>account). >> >>How can we estimate this pure SNR, when we have got real time received >>signal? >> >>Thanking you >> >>Chintan >> > >Hi Chintan, > >one usually knows the power of the transmited signal as well as thepower>of the additive noise at the receiver. Estimation can be done once adata>model is formulated relating the transmited and received signals. Thedata>model can be simple to complicated, depending on how many phenomenaacounts>for. For example, in some cases an appropriate model for the channel isan>FIR filter h[n]. If the transmited signal is s[n] and the additive noiseat>the receiver is w[n], then the data model is >y[n] = conv(h[n],s[n]) + w[n], where y[n] is the received signal. > >Manolis >%%%%% Hi Manolis Thank you very much for your reply. I am aware of this model and we normally use this in simulations. But now lets say in an application of turbo codes in underwater communications. Now for turbo code we need to design 'soft demodulator' where the i/p is received complex symbol and the output will be real valued numbers. But to design this soft demapper, you the noise variance as well (mentioned in a paper by Bauch, 'optimised symbol mappings for BICM-ID'). So how do u find this noise variance in real time or how do u find SNR? Generally, for my simulation I use SNR=Eb/(2*R*sigma^2), where R is code rate. So I use the sigma to scale my noise vector as well as i/p to the soft demapper. Thanks Chintan
Reply by ●June 2, 20082008-06-02
On Mon, 02 Jun 2008 16:04:00 -0500, "cpshah99" <cpshah99@rediffmail.com> wrote:>>>Hello People >>> >>>I have one question to ask. >>> >>>In real time environment, we face multipath, doppler, noise etc. >>> >>>Now when we perform simulations, we plot BER vs SNR plot. In that the >>SNR >>>is pure SNR (pure in the sense SNR=Eb/No, where we just take noise into >>>account). >>> >>>How can we estimate this pure SNR, when we have got real time received >>>signal? >>> >>>Thanking you >>> >>>Chintan >>> >> >>Hi Chintan, >> >>one usually knows the power of the transmited signal as well as the >power >>of the additive noise at the receiver. Estimation can be done once a >data >>model is formulated relating the transmited and received signals. The >data >>model can be simple to complicated, depending on how many phenomena >acounts >>for. For example, in some cases an appropriate model for the channel is >an >>FIR filter h[n]. If the transmited signal is s[n] and the additive noise >at >>the receiver is w[n], then the data model is >>y[n] = conv(h[n],s[n]) + w[n], where y[n] is the received signal. >> >>Manolis >> >%%%%% > >Hi Manolis > >Thank you very much for your reply. I am aware of this model and we >normally use this in simulations. > >But now lets say in an application of turbo codes in underwater >communications. Now for turbo code we need to design 'soft demodulator' >where the i/p is received complex symbol and the output will be real valued >numbers. But to design this soft demapper, you the noise variance as well >(mentioned in a paper by Bauch, 'optimised symbol mappings for BICM-ID'). > >So how do u find this noise variance in real time or how do u find SNR? > >Generally, for my simulation I use SNR=Eb/(2*R*sigma^2), where R is code >rate. So I use the sigma to scale my noise vector as well as i/p to the >soft demapper. > >Thanks > >ChintanIn practice many Turbo Decoders work well with the SNR parameter set to the lowest expected SNR and just left there. This does provide a small performance degradation at higher SNRs but it is usually negligible. Otherwise, SNR can be computed using the slicer in the demodulator and measuring the error vector for each received symbol. Another method is to estimate the BER and then compute the SNR that matches that BER. There are many methods, I think I've seen all of the above used before, but what works best for you may depend on your system and your design constraints. Eric Jacobsen Minister of Algorithms Abineau Communications http://www.ericjacobsen.org Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
Reply by ●June 2, 20082008-06-02
>On Mon, 02 Jun 2008 16:04:00 -0500, "cpshah99" ><cpshah99@rediffmail.com> wrote: > >>>>Hello People >>>> >>>>I have one question to ask. >>>> >>>>In real time environment, we face multipath, doppler, noise etc. >>>> >>>>Now when we perform simulations, we plot BER vs SNR plot. In that the >>>SNR >>>>is pure SNR (pure in the sense SNR=Eb/No, where we just take noiseinto>>>>account). >>>> >>>>How can we estimate this pure SNR, when we have got real timereceived>>>>signal? >>>> >>>>Thanking you >>>> >>>>Chintan >>>> >>> >>>Hi Chintan, >>> >>>one usually knows the power of the transmited signal as well as the >>power >>>of the additive noise at the receiver. Estimation can be done once a >>data >>>model is formulated relating the transmited and received signals. The >>data >>>model can be simple to complicated, depending on how many phenomena >>acounts >>>for. For example, in some cases an appropriate model for the channelis>>an >>>FIR filter h[n]. If the transmited signal is s[n] and the additivenoise>>at >>>the receiver is w[n], then the data model is >>>y[n] = conv(h[n],s[n]) + w[n], where y[n] is the received signal. >>> >>>Manolis >>> >>%%%%% >> >>Hi Manolis >> >>Thank you very much for your reply. I am aware of this model and we >>normally use this in simulations. >> >>But now lets say in an application of turbo codes in underwater >>communications. Now for turbo code we need to design 'soft demodulator' >>where the i/p is received complex symbol and the output will be realvalued>>numbers. But to design this soft demapper, you the noise variance aswell>>(mentioned in a paper by Bauch, 'optimised symbol mappings forBICM-ID').>> >>So how do u find this noise variance in real time or how do u find SNR? >> >>Generally, for my simulation I use SNR=Eb/(2*R*sigma^2), where R iscode>>rate. So I use the sigma to scale my noise vector as well as i/p to the >>soft demapper. >> >>Thanks >> >>Chintan > >In practice many Turbo Decoders work well with the SNR parameter set >to the lowest expected SNR and just left there. This does provide a >small performance degradation at higher SNRs but it is usually >negligible. > >Otherwise, SNR can be computed using the slicer in the demodulator and >measuring the error vector for each received symbol. > >Another method is to estimate the BER and then compute the SNR that >matches that BER. > >There are many methods, I think I've seen all of the above used >before, but what works best for you may depend on your system and your >design constraints. > >Eric Jacobsen >Minister of Algorithms >Abineau Communications >http://www.ericjacobsen.org > >Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php >%%%% Hi Eric Thank you very much. I have read somewhere that in underwater communication, the SNR is rarely above 15 dB. So should I calculate sigma using this value and supply it to demapper? And the other method about using slicer in demodulator and calculating error, if I am not wrong is to calculate SINR (Signal to interference noise ratio). i.e. SINR(dB)=10*log10(Es/sum(error)); where error=S_est - S; where S could be known training sequence or quantised symbol and S_est is o/p of equaliser. And use this SINR to calculate sigma. During one discussion somebody told me to find rms value of the received packet and rms value of silent period( i.e. just noise) and take ratio of it? Does this sound correct? Thanking you. Chintan
Reply by ●June 2, 20082008-06-02
On Jun 3, 5:23 am, "cpshah99" <cpsha...@rediffmail.com> wrote:> >On Mon, 02 Jun 2008 16:04:00 -0500, "cpshah99" > ><cpsha...@rediffmail.com> wrote: > > >>>>Hello People > > >>>>I have one question to ask. > > >>>>In real time environment, we face multipath, doppler, noise etc. > > >>>>Now when we perform simulations, we plot BER vs SNR plot. In that the > >>>SNR > >>>>is pure SNR (pure in the sense SNR=Eb/No, where we just take noise > into > >>>>account). > > >>>>How can we estimate this pure SNR, when we have got real time > received > >>>>signal? > > >>>>Thanking you > > >>>>Chintan > > >>>Hi Chintan, > > >>>one usually knows the power of the transmited signal as well as the > >>power > >>>of the additive noise at the receiver. Estimation can be done once a > >>data > >>>model is formulated relating the transmited and received signals. The > >>data > >>>model can be simple to complicated, depending on how many phenomena > >>acounts > >>>for. For example, in some cases an appropriate model for the channel > is > >>an > >>>FIR filter h[n]. If the transmited signal is s[n] and the additive > noise > >>at > >>>the receiver is w[n], then the data model is > >>>y[n] = conv(h[n],s[n]) + w[n], where y[n] is the received signal. > > >>>Manolis > > >>%%%%% > > >>Hi Manolis > > >>Thank you very much for your reply. I am aware of this model and we > >>normally use this in simulations. > > >>But now lets say in an application of turbo codes in underwater > >>communications. Now for turbo code we need to design 'soft demodulator' > >>where the i/p is received complex symbol and the output will be real > valued > >>numbers. But to design this soft demapper, you the noise variance as > well > >>(mentioned in a paper by Bauch, 'optimised symbol mappings for > BICM-ID'). > > >>So how do u find this noise variance in real time or how do u find SNR? > > >>Generally, for my simulation I use SNR=Eb/(2*R*sigma^2), where R is > code > >>rate. So I use the sigma to scale my noise vector as well as i/p to the > >>soft demapper. > > >>Thanks > > >>Chintan > > >In practice many Turbo Decoders work well with the SNR parameter set > >to the lowest expected SNR and just left there. This does provide a > >small performance degradation at higher SNRs but it is usually > >negligible. > > >Otherwise, SNR can be computed using the slicer in the demodulator and > >measuring the error vector for each received symbol. > > >Another method is to estimate the BER and then compute the SNR that > >matches that BER. > > >There are many methods, I think I've seen all of the above used > >before, but what works best for you may depend on your system and your > >design constraints. > > >Eric Jacobsen > >Minister of Algorithms > >Abineau Communications > >http://www.ericjacobsen.org > > >Blog:http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php > > %%%% > > Hi Eric > > Thank you very much. I have read somewhere that in underwater > communication, the SNR is rarely above 15 dB. So should I calculate sigma > using this value and supply it to demapper? > > And the other method about using slicer in demodulator and calculating > error, if I am not wrong is to calculate SINR (Signal to interference noise > ratio). i.e. > > SINR(dB)=10*log10(Es/sum(error)); > > where error=S_est - S; > where S could be known training sequence or quantised symbol and S_est is > o/p of equaliser. > > And use this SINR to calculate sigma. > > During one discussion somebody told me to find rms value of the received > packet and rms value of silent period( i.e. just noise) and take ratio of > it? Does this sound correct?Just thinking off the top of my head, someone please correct me if I'm wrong - Wouldn't the RMS value of the received packet include noise in it as well? Which would mean that the estimate you get is actually 1 + SNR?
Reply by ●June 3, 20082008-06-03
Karthik wrote:> On Jun 3, 5:23 am, "cpshah99" <cpsha...@rediffmail.com> wrote: >>> On Mon, 02 Jun 2008 16:04:00 -0500, "cpshah99" >>> <cpsha...@rediffmail.com> wrote: >>>>>> Hello People >>>>>> I have one question to ask. >>>>>> In real time environment, we face multipath, doppler, noise etc. >>>>>> Now when we perform simulations, we plot BER vs SNR plot. In that the >>>>> SNR >>>>>> is pure SNR (pure in the sense SNR=Eb/No, where we just take noise >> into >>>>>> account). >>>>>> How can we estimate this pure SNR, when we have got real time >> received >>>>>> signal? >>>>>> Thanking you >>>>>> Chintan >>>>> Hi Chintan, >>>>> one usually knows the power of the transmited signal as well as the >>>> power >>>>> of the additive noise at the receiver. Estimation can be done once a >>>> data >>>>> model is formulated relating the transmited and received signals. The >>>> data >>>>> model can be simple to complicated, depending on how many phenomena >>>> acounts >>>>> for. For example, in some cases an appropriate model for the channel >> is >>>> an >>>>> FIR filter h[n]. If the transmited signal is s[n] and the additive >> noise >>>> at >>>>> the receiver is w[n], then the data model is >>>>> y[n] = conv(h[n],s[n]) + w[n], where y[n] is the received signal. >>>>> Manolis >>>> %%%%% >>>> Hi Manolis >>>> Thank you very much for your reply. I am aware of this model and we >>>> normally use this in simulations. >>>> But now lets say in an application of turbo codes in underwater >>>> communications. Now for turbo code we need to design 'soft demodulator' >>>> where the i/p is received complex symbol and the output will be real >> valued >>>> numbers. But to design this soft demapper, you the noise variance as >> well >>>> (mentioned in a paper by Bauch, 'optimised symbol mappings for >> BICM-ID'). >> >>>> So how do u find this noise variance in real time or how do u find SNR? >>>> Generally, for my simulation I use SNR=Eb/(2*R*sigma^2), where R is >> code >>>> rate. So I use the sigma to scale my noise vector as well as i/p to the >>>> soft demapper. >>>> Thanks >>>> Chintan >>> In practice many Turbo Decoders work well with the SNR parameter set >>> to the lowest expected SNR and just left there. This does provide a >>> small performance degradation at higher SNRs but it is usually >>> negligible. >>> Otherwise, SNR can be computed using the slicer in the demodulator and >>> measuring the error vector for each received symbol. >>> Another method is to estimate the BER and then compute the SNR that >>> matches that BER. >>> There are many methods, I think I've seen all of the above used >>> before, but what works best for you may depend on your system and your >>> design constraints. >>> Eric Jacobsen >>> Minister of Algorithms >>> Abineau Communications >>> http://www.ericjacobsen.org >>> Blog:http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php >> %%%% >> >> Hi Eric >> >> Thank you very much. I have read somewhere that in underwater >> communication, the SNR is rarely above 15 dB. So should I calculate sigma >> using this value and supply it to demapper? >> >> And the other method about using slicer in demodulator and calculating >> error, if I am not wrong is to calculate SINR (Signal to interference noise >> ratio). i.e. >> >> SINR(dB)=10*log10(Es/sum(error)); >> >> where error=S_est - S; >> where S could be known training sequence or quantised symbol and S_est is >> o/p of equaliser. >> >> And use this SINR to calculate sigma. >> >> During one discussion somebody told me to find rms value of the received >> packet and rms value of silent period( i.e. just noise) and take ratio of >> it? Does this sound correct? > > Just thinking off the top of my head, someone please correct me if I'm > wrong - > > Wouldn't the RMS value of the received packet include noise in it as > well? Which > would mean that the estimate you get is actually 1 + SNR?If the SNR is at all reasonable, the correction will be lost in the uncertainty of the measurement. The signals powers add. Forget noise for a moment. Assume a power level of 0 dBm, to which you add a signal of -20 dBm. Assume also no constructive or destructive interference. What is the new signal level? Hint: less than 0.05 dBm Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●June 3, 20082008-06-03
On Jun 3, 8:01 am, Jerry Avins <j...@ieee.org> wrote:> Karthik wrote: > > On Jun 3, 5:23 am, "cpshah99" <cpsha...@rediffmail.com> wrote: > >>> On Mon, 02 Jun 2008 16:04:00 -0500, "cpshah99" > >>> <cpsha...@rediffmail.com> wrote: > >>>>>> Hello People > >>>>>> I have one question to ask. > >>>>>> In real time environment, we face multipath, doppler, noise etc. > >>>>>> Now when we perform simulations, we plot BER vs SNR plot. In that the > >>>>> SNR > >>>>>> is pure SNR (pure in the sense SNR=Eb/No, where we just take noise > >> into > >>>>>> account). > >>>>>> How can we estimate this pure SNR, when we have got real time > >> received > >>>>>> signal? > >>>>>> Thanking you > >>>>>> Chintan > >>>>> Hi Chintan, > >>>>> one usually knows the power of the transmited signal as well as the > >>>> power > >>>>> of the additive noise at the receiver. Estimation can be done once a > >>>> data > >>>>> model is formulated relating the transmited and received signals. The > >>>> data > >>>>> model can be simple to complicated, depending on how many phenomena > >>>> acounts > >>>>> for. For example, in some cases an appropriate model for the channel > >> is > >>>> an > >>>>> FIR filter h[n]. If the transmited signal is s[n] and the additive > >> noise > >>>> at > >>>>> the receiver is w[n], then the data model is > >>>>> y[n] = conv(h[n],s[n]) + w[n], where y[n] is the received signal. > >>>>> Manolis > >>>> %%%%% > >>>> Hi Manolis > >>>> Thank you very much for your reply. I am aware of this model and we > >>>> normally use this in simulations. > >>>> But now lets say in an application of turbo codes in underwater > >>>> communications. Now for turbo code we need to design 'soft demodulator' > >>>> where the i/p is received complex symbol and the output will be real > >> valued > >>>> numbers. But to design this soft demapper, you the noise variance as > >> well > >>>> (mentioned in a paper by Bauch, 'optimised symbol mappings for > >> BICM-ID'). > > >>>> So how do u find this noise variance in real time or how do u find SNR? > >>>> Generally, for my simulation I use SNR=Eb/(2*R*sigma^2), where R is > >> code > >>>> rate. So I use the sigma to scale my noise vector as well as i/p to the > >>>> soft demapper. > >>>> Thanks > >>>> Chintan > >>> In practice many Turbo Decoders work well with the SNR parameter set > >>> to the lowest expected SNR and just left there. This does provide a > >>> small performance degradation at higher SNRs but it is usually > >>> negligible. > >>> Otherwise, SNR can be computed using the slicer in the demodulator and > >>> measuring the error vector for each received symbol. > >>> Another method is to estimate the BER and then compute the SNR that > >>> matches that BER. > >>> There are many methods, I think I've seen all of the above used > >>> before, but what works best for you may depend on your system and your > >>> design constraints. > >>> Eric Jacobsen > >>> Minister of Algorithms > >>> Abineau Communications > >>>http://www.ericjacobsen.org > >>> Blog:http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php > >> %%%% > > >> Hi Eric > > >> Thank you very much. I have read somewhere that in underwater > >> communication, the SNR is rarely above 15 dB. So should I calculate sigma > >> using this value and supply it to demapper? > > >> And the other method about using slicer in demodulator and calculating > >> error, if I am not wrong is to calculate SINR (Signal to interference noise > >> ratio). i.e. > > >> SINR(dB)=10*log10(Es/sum(error)); > > >> where error=S_est - S; > >> where S could be known training sequence or quantised symbol and S_est is > >> o/p of equaliser. > > >> And use this SINR to calculate sigma. > > >> During one discussion somebody told me to find rms value of the received > >> packet and rms value of silent period( i.e. just noise) and take ratio of > >> it? Does this sound correct? > > > Just thinking off the top of my head, someone please correct me if I'm > > wrong - > > > Wouldn't the RMS value of the received packet include noise in it as > > well? Which > > would mean that the estimate you get is actually 1 + SNR? > > If the SNR is at all reasonable, the correction will be lost in the > uncertainty of the measurement. The signals powers add. Forget noise for > a moment. Assume a power level of 0 dBm, to which you add a signal of > -20 dBm. Assume also no constructive or destructive interference. What > is the new signal level? Hint: less than 0.05 dBm > > Jerry > -- > Engineering is the art of making what you want from things you can get. > �����������������������������������������������������������������������Jerry, what I understand from your post is that if the signal and noise powers are widely separated, the estimate given by (RMS_recd_pkt / RMS_silent) is a good estimate of the SNR. However, if the signal and noise powers are comparable, it looks like this estimate would be off by quite a bit. If I understand correctly, it shouldn't make much of a difference because the SNR is pretty bad anyway, and nothing can be done. Is that right or did I miss something? BTW, a little googling seems to show that some folks actually *define* the SNR to be P_recd / P_noise, in which case my objection vanishes into thin air. What's the correct way of defining the SNR?
Reply by ●June 3, 20082008-06-03
On Jun 3, 8:01 am, Jerry Avins <j...@ieee.org> wrote:> Karthik wrote: > > On Jun 3, 5:23 am, "cpshah99" <cpsha...@rediffmail.com> wrote: > >>> On Mon, 02 Jun 2008 16:04:00 -0500, "cpshah99" > >>> <cpsha...@rediffmail.com> wrote: > >>>>>> Hello People > >>>>>> I have one question to ask. > >>>>>> In real time environment, we face multipath, doppler, noise etc. > >>>>>> Now when we perform simulations, we plot BER vs SNR plot. In that the > >>>>> SNR > >>>>>> is pure SNR (pure in the sense SNR=Eb/No, where we just take noise > >> into > >>>>>> account). > >>>>>> How can we estimate this pure SNR, when we have got real time > >> received > >>>>>> signal? > >>>>>> Thanking you > >>>>>> Chintan > >>>>> Hi Chintan, > >>>>> one usually knows the power of the transmited signal as well as the > >>>> power > >>>>> of the additive noise at the receiver. Estimation can be done once a > >>>> data > >>>>> model is formulated relating the transmited and received signals. The > >>>> data > >>>>> model can be simple to complicated, depending on how many phenomena > >>>> acounts > >>>>> for. For example, in some cases an appropriate model for the channel > >> is > >>>> an > >>>>> FIR filter h[n]. If the transmited signal is s[n] and the additive > >> noise > >>>> at > >>>>> the receiver is w[n], then the data model is > >>>>> y[n] = conv(h[n],s[n]) + w[n], where y[n] is the received signal. > >>>>> Manolis > >>>> %%%%% > >>>> Hi Manolis > >>>> Thank you very much for your reply. I am aware of this model and we > >>>> normally use this in simulations. > >>>> But now lets say in an application of turbo codes in underwater > >>>> communications. Now for turbo code we need to design 'soft demodulator' > >>>> where the i/p is received complex symbol and the output will be real > >> valued > >>>> numbers. But to design this soft demapper, you the noise variance as > >> well > >>>> (mentioned in a paper by Bauch, 'optimised symbol mappings for > >> BICM-ID'). > > >>>> So how do u find this noise variance in real time or how do u find SNR? > >>>> Generally, for my simulation I use SNR=Eb/(2*R*sigma^2), where R is > >> code > >>>> rate. So I use the sigma to scale my noise vector as well as i/p to the > >>>> soft demapper. > >>>> Thanks > >>>> Chintan > >>> In practice many Turbo Decoders work well with the SNR parameter set > >>> to the lowest expected SNR and just left there. This does provide a > >>> small performance degradation at higher SNRs but it is usually > >>> negligible. > >>> Otherwise, SNR can be computed using the slicer in the demodulator and > >>> measuring the error vector for each received symbol. > >>> Another method is to estimate the BER and then compute the SNR that > >>> matches that BER. > >>> There are many methods, I think I've seen all of the above used > >>> before, but what works best for you may depend on your system and your > >>> design constraints. > >>> Eric Jacobsen > >>> Minister of Algorithms > >>> Abineau Communications > >>>http://www.ericjacobsen.org > >>> Blog:http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php > >> %%%% > > >> Hi Eric > > >> Thank you very much. I have read somewhere that in underwater > >> communication, the SNR is rarely above 15 dB. So should I calculate sigma > >> using this value and supply it to demapper? > > >> And the other method about using slicer in demodulator and calculating > >> error, if I am not wrong is to calculate SINR (Signal to interference noise > >> ratio). i.e. > > >> SINR(dB)=10*log10(Es/sum(error)); > > >> where error=S_est - S; > >> where S could be known training sequence or quantised symbol and S_est is > >> o/p of equaliser. > > >> And use this SINR to calculate sigma. > > >> During one discussion somebody told me to find rms value of the received > >> packet and rms value of silent period( i.e. just noise) and take ratio of > >> it? Does this sound correct? > > > Just thinking off the top of my head, someone please correct me if I'm > > wrong - > > > Wouldn't the RMS value of the received packet include noise in it as > > well? Which > > would mean that the estimate you get is actually 1 + SNR? > > If the SNR is at all reasonable, the correction will be lost in the > uncertainty of the measurement. The signals powers add. Forget noise for > a moment. Assume a power level of 0 dBm, to which you add a signal of > -20 dBm. Assume also no constructive or destructive interference. What > is the new signal level? Hint: less than 0.05 dBm > > Jerry > -- > Engineering is the art of making what you want from things you can get. > �����������������������������������������������������������������������Not sure if my last post got through, google seems to have swallowed it. Jerry, if the signal and noise powers are widely separated, then the estimate given by P_recd_pkt / P_silent is pretty close to the "actual" SNR. However, if the signal and noise powers are of comparable magnitude, then this is no longer true - it doesn't / shouldn't matter much, because the SNR is so bad that nothing can be done to recover the signal. Is that right, or did I miss something? I have a more basic question as well :) Let's say we have channel modelled by an FIR filter h(n), transmitted signal x(n), received signal y(n) and noise w(n). y(n) = x(n) * h(n) + w(n). Some folks define the SNR at the receiving end to be RMS_(y - w) / RMS_w, and other folks define it as RMS_y / RMS_w. I was under the impression that it was the first, but now I'm confused - which is the correct way to define SNR?
Reply by ●June 3, 20082008-06-03
Karthik wrote: ...> Jerry, what I understand from your post is that if the signal and > noise powers are widely > separated, the estimate given by (RMS_recd_pkt / RMS_silent) is a good > estimate of the SNR.If the signal has ten times the power of the noise, then the difference between S/N and (S+N)/N is less than half a dB.> However, if the signal and noise powers are comparable, it looks like > this estimate would be > off by quite a bit.When signal and noise have equal power, the difference is 3 dB.> If I understand correctly, it shouldn't make much > of a difference because the SNR > is pretty bad anyway, and nothing can be done. > > Is that right or did I miss something?It looks right to me.> BTW, a little googling seems to show that some folks actually *define* > the SNR to be P_recd / P_noise, > in which case my objection vanishes into thin air. What's the correct > way of defining the SNR?As I see it, it doesn't matter much, so it isn't nailed down. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
