On Fri, 01 May 2015 14:50:08 -0500, "ahmed1984" <105515@DSPRelated> wrote:>>>Hi, >>>I need a help please and an answer for this: >>>I'm trying to do a BER simulation for QPSK in OFDM system over a >>>Multipath >>>Channel in Matlab and I'm using the function (ht=rayleighchan()) for >>>model the multipath channel with the following parameters: >>>tau=[0 1e-4 2e-4] second %Path Delay >>>Ts -5 sec or 1e-3 sec(for another simulation) %Sample >>>Period >>>Fd=0; %Maximum Doppler Frequency Shift (Assuming >>>Time-Invariant Channel) >>>PathGains=[0 -2 -5] dB >>> and I transmit my modulated data by using: >>>tx=filter( ht,xt) where the xt is my modulated data. >>>Tx=tx+nt where nt is the AWGN. >>> >>>at the receiver I don't know how to remove the effect of the (ht),i.e, >>>the >>>effect of multipath channel. >>>So please can any one explain how remove this affect or anyone have a >>>matlab code to do so. >>>Thanks in Advance. >>> >>> >>>--------------------------------------- >>>Posted through http://www.DSPRelated.com >> >>You need to get few points right. Forget about matlab for now. >> >>use pen and paper. >> >>The system model is as below in time domain >> >>y = h*x (ignore noise for now, where * is convolution) >> >>Now, if you were to take FFT on both sides, then in the Freq domain, it >>becomes >> >>Y = H.X (where . is now multiplication) >> >>Assuming that you know H at receiver, you can estimate X_est = Y/H >> >>Now, you need to ask yourself, in the absence of noise and under the >>assumption that you know H at receiver, what should be the BER? >> >>I will tell you more, once you tell me the answer of above question. >> >>(Multipath channels are not easy to understand, certainly not by using >>inbuilt function of Matlab. This further annoys me that most of the >>people >>think by using OFDM they can get away with multipath, but what you need >>to >>keep in mind that the channel always works in time domain and it does >>not >>know whether your data is in frequency domain or time domain). >> >>Take your time and study this topic properly, there is not short cut. >> >>Best Regards >> >>Chintan >> >> >> >>--------------------------------------- >>Posted through http://www.DSPRelated.com > >Hi Chintan, >Thank you for your reply. I think in case we already know the channel >impulse response then the BER will be 0 no errors in the absence of the >AWGN.Ideally, yes. In practice channel estimation is never perfect and if the channel is aggressive it may be difficult to completely recover (e.g., channel nulls wipe out a few subcarriers). FEC can often correct for these, so it may depend on whether you're measuring BER before or after FEC. Eric Jacobsen Anchor Hill Communications http://www.anchorhill.com
BER for QPSK in OFDM over Multipath Channel
Started by ●April 30, 2015
Reply by ●May 1, 20152015-05-01
Reply by ●May 2, 20152015-05-02
>>>Hi, >>>I need a help please and an answer for this: >>>I'm trying to do a BER simulation for QPSK in OFDM system over a >>>Multipath >>>Channel in Matlab and I'm using the function (ht=rayleighchan()) for >>>model the multipath channel with the following parameters: >>>tau=[0 1e-4 2e-4] second %Path Delay >>>Ts -5 sec or 1e-3 sec(for another simulation) %Sample >>>Period >>>Fd=0; %Maximum Doppler Frequency Shift (Assuming >>>Time-Invariant Channel) >>>PathGains=[0 -2 -5] dB >>> and I transmit my modulated data by using: >>>tx=filter( ht,xt) where the xt is my modulated data. >>>Tx=tx+nt where nt is the AWGN. >>> >>>at the receiver I don't know how to remove the effect of the >(ht),i.e, >>>the >>>effect of multipath channel. >>>So please can any one explain how remove this affect or anyone have a >>>matlab code to do so. >>>Thanks in Advance. >>> >>> >>>--------------------------------------- >>>Posted through http://www.DSPRelated.com >> >>You need to get few points right. Forget about matlab for now. >> >>use pen and paper. >> >>The system model is as below in time domain >> >>y = h*x (ignore noise for now, where * is convolution) >> >>Now, if you were to take FFT on both sides, then in the Freq domain, >it >>becomes >> >>Y = H.X (where . is now multiplication) >> >>Assuming that you know H at receiver, you can estimate X_est = Y/H >> >>Now, you need to ask yourself, in the absence of noise and under the >>assumption that you know H at receiver, what should be the BER? >> >>I will tell you more, once you tell me the answer of above question. >> >>(Multipath channels are not easy to understand, certainly not by using >>inbuilt function of Matlab. This further annoys me that most of the >>people >>think by using OFDM they can get away with multipath, but what you >need >>to >>keep in mind that the channel always works in time domain and it does >>not >>know whether your data is in frequency domain or time domain). >> >>Take your time and study this topic properly, there is not short cut. >> >>Best Regards >> >>Chintan >> >> >> >>--------------------------------------- >>Posted through http://www.DSPRelated.com > >Hi Chintan, >Thank you for your reply. I think in case we already know the channel >impulse response then the BER will be 0 no errors in the absence of the >AWGN. >--------------------------------------- >Posted through http://www.DSPRelated.comHave a look at this link http://www.dsplog.com/2008/08/26/ofdm-rayleigh-channel-ber-bpsk/ it may help. Chintan --------------------------------------- Posted through http://www.DSPRelated.com
Reply by ●May 2, 20152015-05-02
>On Fri, 01 May 2015 14:50:08 -0500, "ahmed1984" <105515@DSPRelated> >wrote: > >>>>Hi, >>>>I need a help please and an answer for this: >>>>I'm trying to do a BER simulation for QPSK in OFDM system over a >>>>Multipath >>>>Channel in Matlab and I'm using the function (ht=rayleighchan()) >for >>>>model the multipath channel with the following parameters: >>>>tau=[0 1e-4 2e-4] second %Path Delay >>>>Ts -5 sec or 1e-3 sec(for another simulation) %Sample >>>>Period >>>>Fd=0; %Maximum Doppler Frequency Shift (Assuming >>>>Time-Invariant Channel) >>>>PathGains=[0 -2 -5] dB >>>> and I transmit my modulated data by using: >>>>tx=filter( ht,xt) where the xt is my modulated data. >>>>Tx=tx+nt where nt is the AWGN. >>>> >>>>at the receiver I don't know how to remove the effect of the >(ht),i.e, >>>>the >>>>effect of multipath channel. >>>>So please can any one explain how remove this affect or anyone have >a >>>>matlab code to do so. >>>>Thanks in Advance. >>>> >>>> >>>>--------------------------------------- >>>>Posted through http://www.DSPRelated.com >>> >>>You need to get few points right. Forget about matlab for now. >>> >>>use pen and paper. >>> >>>The system model is as below in time domain >>> >>>y = h*x (ignore noise for now, where * is convolution) >>> >>>Now, if you were to take FFT on both sides, then in the Freq domain, >it >>>becomes >>> >>>Y = H.X (where . is now multiplication) >>> >>>Assuming that you know H at receiver, you can estimate X_est = Y/H >>> >>>Now, you need to ask yourself, in the absence of noise and under the >>>assumption that you know H at receiver, what should be the BER? >>> >>>I will tell you more, once you tell me the answer of above question. >>> >>>(Multipath channels are not easy to understand, certainly not by >using >>>inbuilt function of Matlab. This further annoys me that most of the >>>people >>>think by using OFDM they can get away with multipath, but what you >need >>>to >>>keep in mind that the channel always works in time domain and it does >>>not >>>know whether your data is in frequency domain or time domain). >>> >>>Take your time and study this topic properly, there is not short cut. >>> >>>Best Regards >>> >>>Chintan >>> >>> >>> >>>--------------------------------------- >>>Posted through http://www.DSPRelated.com >> >>Hi Chintan, >>Thank you for your reply. I think in case we already know the channel >>impulse response then the BER will be 0 no errors in the absence of >the >>AWGN. > >Ideally, yes. In practice channel estimation is never perfect and if >the channel is aggressive it may be difficult to completely recover >(e.g., channel nulls wipe out a few subcarriers). FEC can often >correct for these, so it may depend on whether you're measuring BER >before or after FEC. > > >Eric Jacobsen >Anchor Hill Communications >http://www.anchorhill.comHi Eric, I looking for the uncoded BER curve so let say before the FEC. Thanks --------------------------------------- Posted through http://www.DSPRelated.com
Reply by ●May 2, 20152015-05-02
>>>>Hi, >>>>I need a help please and an answer for this: >>>>I'm trying to do a BER simulation for QPSK in OFDM system over a >>>>Multipath >>>>Channel in Matlab and I'm using the function (ht=rayleighchan()) >for >>>>model the multipath channel with the following parameters: >>>>tau=[0 1e-4 2e-4] second %Path Delay >>>>Ts -5 sec or 1e-3 sec(for another simulation) %Sample >>>>Period >>>>Fd=0; %Maximum Doppler Frequency Shift (Assuming >>>>Time-Invariant Channel) >>>>PathGains=[0 -2 -5] dB >>>> and I transmit my modulated data by using: >>>>tx=filter( ht,xt) where the xt is my modulated data. >>>>Tx=tx+nt where nt is the AWGN. >>>> >>>>at the receiver I don't know how to remove the effect of the >>(ht),i.e, >>>>the >>>>effect of multipath channel. >>>>So please can any one explain how remove this affect or anyone have >a >>>>matlab code to do so. >>>>Thanks in Advance. >>>> >>>> >>>>--------------------------------------- >>>>Posted through http://www.DSPRelated.com >>> >>>You need to get few points right. Forget about matlab for now. >>> >>>use pen and paper. >>> >>>The system model is as below in time domain >>> >>>y = h*x (ignore noise for now, where * is convolution) >>> >>>Now, if you were to take FFT on both sides, then in the Freq domain, >>it >>>becomes >>> >>>Y = H.X (where . is now multiplication) >>> >>>Assuming that you know H at receiver, you can estimate X_est = Y/H >>> >>>Now, you need to ask yourself, in the absence of noise and under the >>>assumption that you know H at receiver, what should be the BER? >>> >>>I will tell you more, once you tell me the answer of above question. >>> >>>(Multipath channels are not easy to understand, certainly not by >using >>>inbuilt function of Matlab. This further annoys me that most of the >>>people >>>think by using OFDM they can get away with multipath, but what you >>need >>>to >>>keep in mind that the channel always works in time domain and it does >>>not >>>know whether your data is in frequency domain or time domain). >>> >>>Take your time and study this topic properly, there is not short cut. >>> >>>Best Regards >>> >>>Chintan >>> >>> >>> >>>--------------------------------------- >>>Posted through http://www.DSPRelated.com >> >>Hi Chintan, >>Thank you for your reply. I think in case we already know the channel >>impulse response then the BER will be 0 no errors in the absence of >the >>AWGN. >>--------------------------------------- >>Posted through http://www.DSPRelated.com > >Have a look at this link >http://www.dsplog.com/2008/08/26/ofdm-rayleigh-channel-ber-bpsk/ > >it may help. > >Chintan >--------------------------------------- >Posted through http://www.DSPRelated.comHi Chintan, Thank you for the help and this link is really helpful I knew it but didn't figure out how he dealt with his channel model I can't find where is the delay of each tap it is only the path gains introduced as I understand it. Thanks --------------------------------------- Posted through http://www.DSPRelated.com
Reply by ●May 2, 20152015-05-02
Eric Jacobsen <eric.jacobsen@ieee.org> wrote:>You have to estimate the channel in the receiver, which can be done >using a variety of different methods, including a preamble or pilot >tones or whatever best fits your system.>Once you have estimated the channel you should have a single >complex-valued correction coefficient for each subcarrier. >Multiplying each subcarrier by this coefficient to remove the effects >of the channel should get you back to having reasonable BER.>In your case you know h(t) already, so you can sample the channel >response in the frequency domain with the same spacing as the >subcarriers and apply those coefficient to equalize the channel.Yes, and I would expand on the "apply those coefficients" part of it. If there is an FEC decoder with a soft-decision input, you use the coefficients to create metrics having the effect of under-weighting the weaker subcarriers. If there is no FEC decoder, then in the baseline case you are kind of in trouble. OFDM makes little sense without FEC -- with no coding (or waterfilling, etc.) the BER curve will be dominated by the SNR associated with weakest subcarriers. Steve
Reply by ●May 3, 20152015-05-03
On Sat, 02 May 2015 16:47:19 -0500, "ahmed1984" <105515@DSPRelated> wrote:>>On Fri, 01 May 2015 14:50:08 -0500, "ahmed1984" <105515@DSPRelated> >>wrote: >> >>>>>Hi, >>>>>I need a help please and an answer for this: >>>>>I'm trying to do a BER simulation for QPSK in OFDM system over a >>>>>Multipath >>>>>Channel in Matlab and I'm using the function (ht=rayleighchan()) >>for >>>>>model the multipath channel with the following parameters: >>>>>tau=[0 1e-4 2e-4] second %Path Delay >>>>>Ts -5 sec or 1e-3 sec(for another simulation) %Sample >>>>>Period >>>>>Fd=0; %Maximum Doppler Frequency Shift (Assuming >>>>>Time-Invariant Channel) >>>>>PathGains=[0 -2 -5] dB >>>>> and I transmit my modulated data by using: >>>>>tx=filter( ht,xt) where the xt is my modulated data. >>>>>Tx=tx+nt where nt is the AWGN. >>>>> >>>>>at the receiver I don't know how to remove the effect of the >>(ht),i.e, >>>>>the >>>>>effect of multipath channel. >>>>>So please can any one explain how remove this affect or anyone have >>a >>>>>matlab code to do so. >>>>>Thanks in Advance. >>>>> >>>>> >>>>>--------------------------------------- >>>>>Posted through http://www.DSPRelated.com >>>> >>>>You need to get few points right. Forget about matlab for now. >>>> >>>>use pen and paper. >>>> >>>>The system model is as below in time domain >>>> >>>>y = h*x (ignore noise for now, where * is convolution) >>>> >>>>Now, if you were to take FFT on both sides, then in the Freq domain, >>it >>>>becomes >>>> >>>>Y = H.X (where . is now multiplication) >>>> >>>>Assuming that you know H at receiver, you can estimate X_est = Y/H >>>> >>>>Now, you need to ask yourself, in the absence of noise and under the >>>>assumption that you know H at receiver, what should be the BER? >>>> >>>>I will tell you more, once you tell me the answer of above question. >>>> >>>>(Multipath channels are not easy to understand, certainly not by >>using >>>>inbuilt function of Matlab. This further annoys me that most of the >>>>people >>>>think by using OFDM they can get away with multipath, but what you >>need >>>>to >>>>keep in mind that the channel always works in time domain and it does >>>>not >>>>know whether your data is in frequency domain or time domain). >>>> >>>>Take your time and study this topic properly, there is not short cut. >>>> >>>>Best Regards >>>> >>>>Chintan >>>> >>>> >>>> >>>>--------------------------------------- >>>>Posted through http://www.DSPRelated.com >>> >>>Hi Chintan, >>>Thank you for your reply. I think in case we already know the channel >>>impulse response then the BER will be 0 no errors in the absence of >>the >>>AWGN. >> >>Ideally, yes. In practice channel estimation is never perfect and if >>the channel is aggressive it may be difficult to completely recover >>(e.g., channel nulls wipe out a few subcarriers). FEC can often >>correct for these, so it may depend on whether you're measuring BER >>before or after FEC. >> >> >>Eric Jacobsen >>Anchor Hill Communications >>http://www.anchorhill.com > >Hi Eric, >I looking for the uncoded BER curve so let say before the FEC. >Thanks >--------------------------------------- >Posted through http://www.DSPRelated.comAs Steve Pope mentioned, OFDM in multipath without FEC kinda sucks, because the badly faded (e.g., nulled) subcarrier information can't be recovered. Even if the SNR of the overall signal is high, the errors from the faded subcarriers will persist. Since many OFDM systems don't have large numbers of data-carrying subcarriers, you can wind up with a BER of much worse than 10e-3 (i.e., bad) from a single nulled subcarrier. Usually if one is nulled, others will be suffering, so it can get pretty horrible even with a very high SNR. The 'C' in COFDM is therefore usually assumed. ;) Nevertheless, there can be good reasons to analyze the BER prior to FEC decoding, so just be aware that it won't be unusual at all for the performance to suck badly when a multipath model is applied, even when the noise power is zero. Eric Jacobsen Anchor Hill Communications http://www.anchorhill.com
Reply by ●May 3, 20152015-05-03
eric.jacobsen@ieee.org (Eric Jacobsen) writes:> [...] > > Nevertheless, there can be good reasons to analyze the BER prior to > FEC decoding, so just be aware that it won't be unusual at all for the > performance to suck badly when a multipath model is applied, even when > the noise power is zero.So where is the noise coming from? Numerical precision? -- Randy Yates Digital Signal Labs http://www.digitalsignallabs.com
Reply by ●May 3, 20152015-05-03
On Sat, 02 May 2015 23:57:43 -0400, Randy Yates <yates@digitalsignallabs.com> wrote:>eric.jacobsen@ieee.org (Eric Jacobsen) writes: >> [...] >> >> Nevertheless, there can be good reasons to analyze the BER prior to >> FEC decoding, so just be aware that it won't be unusual at all for the >> performance to suck badly when a multipath model is applied, even when >> the noise power is zero. > >So where is the noise coming from? Numerical precision?Essentially. A fade can essentially be a spectral null, where a zero in the channel frequency response lands on a subcarrier, so essentially no energy from that subcarrier makes it into the receiver. Likewise a subcarrier can be faded badly enough that whatever precision is being used isn't sufficient for reliable demodulation. Eric Jacobsen Anchor Hill Communications http://www.anchorhill.com
Reply by ●May 3, 20152015-05-03
Eric Jacobsen <eric.jacobsen@ieee.org> wrote:>On Sat, 02 May 2015 23:57:43 -0400, Randy Yates>>eric.jacobsen@ieee.org (Eric Jacobsen) writes:>>> Nevertheless, there can be good reasons to analyze the BER prior to >>> FEC decoding, so just be aware that it won't be unusual at all for the >>> performance to suck badly when a multipath model is applied, even when >>> the noise power is zero.>>So where is the noise coming from? Numerical precision?>Essentially. A fade can essentially be a spectral null, where a zero >in the channel frequency response lands on a subcarrier, so >essentially no energy from that subcarrier makes it into the receiver.>Likewise a subcarrier can be faded badly enough that whatever >precision is being used isn't sufficient for reliable demodulation.I agree with this analysis but I'm not sure how informative such an experiment is. Typically one would design the numerical precision in the receiver such that, with added noise over the expected range, the numerical precision itself does not degrade the operating point by more than some target (say, 0.1 dB). If you remove the added noise, under very weak signal conditions you would observe the degrading effects of your numerical precision. But this experiment itself does not tell you what precision you need. Or most of the time it doesn't. Steve
