Forums

Oversampling in WCDMA receiver

Started by HSDPA-boy August 17, 2006
Hi,

I have a question about oversampling when doing RRC filtering and
correlations in WCDMA receiver. Both these operations are done at base
band in the digital domain. In the literature I've read that the signal is
oversampled say 4 or 6 times when doing both the RRC filtering and
subsequent correlation. What is the motivation for oversampling in these
two cases? The use of an oversampled signal when correlating especially
confuses me. I've always been under the impression that correlation is
done with one sample per chip. Hope you guys can help me understand
this...


In article <Bo6dnbodSYaPwHnZnZ2dnUVZ_o6dnZ2d@giganews.com>,
 "HSDPA-boy" <fearsome.green@gmail.com> wrote:

|  Hi,
|  
|  I have a question about oversampling when doing RRC filtering and
|  correlations in WCDMA receiver. Both these operations are done at base
|  band in the digital domain. In the literature I've read that the signal is
|  oversampled say 4 or 6 times when doing both the RRC filtering and
|  subsequent correlation. What is the motivation for oversampling in these
|  two cases?

Multipath delays aren't quantized to the chip rate, so it is beneficial 
to oversample at the receiver to better center the individual rake 
delays to match peak correlation for a given mulitpath.  For example, if 
a multipath delay is exactly a multiple of 1/2 the chip rate, then 
oversampling by 2x would give a 30% gain to the detected peak.

To help minimize inter-symbol interference when doing this, the 
transmitted data is oversampled by the  same amount, and filtered 
through a root-raised cosine filter at both ends.

   -- Tim Olson
>In article <Bo6dnbodSYaPwHnZnZ2dnUVZ_o6dnZ2d@giganews.com>, > "HSDPA-boy" <fearsome.green@gmail.com> wrote: > >| Hi, >| >| I have a question about oversampling when doing RRC filtering and >| correlations in WCDMA receiver. Both these operations are done at
base
>| band in the digital domain. In the literature I've read that the
signal is
>| oversampled say 4 or 6 times when doing both the RRC filtering and >| subsequent correlation. What is the motivation for oversampling in
these
>| two cases? > >Multipath delays aren't quantized to the chip rate, so it is beneficial >to oversample at the receiver to better center the individual rake >delays to match peak correlation for a given mulitpath. For example, if
>a multipath delay is exactly a multiple of 1/2 the chip rate, then >oversampling by 2x would give a 30% gain to the detected peak. > >To help minimize inter-symbol interference when doing this, the >transmitted data is oversampled by the same amount, and filtered >through a root-raised cosine filter at both ends. > > -- Tim Olson >
Thanks for the reply, Tim. According to theory the time resolution at the receiver is approximately one over the system bandwidth. Since the bandwidth of the RRC filter is actually a bit larger than one over the symbol (chip) duration I can see why an oversampling of 2x would make sense. But is useful to oversample even more? I mean, two multipaths that are separated by less than half of the chip duration would appear as a single multipath after RRC filtering. Or am I wrong in my reasoning? Looking forward to your thoughts. /Fred
In article <Z8GdnXa8NuBaP3fZnZ2dnUVZ_v2dnZ2d@giganews.com>,
 "HSDPA-boy" <fearsome.green@gmail.com> wrote:


|  According to theory the time resolution at the receiver is approximately
|  one over the system bandwidth. Since the bandwidth of the RRC filter is
|  actually a bit larger than one over the symbol (chip) duration I can see
|  why an oversampling of 2x would make sense. But is useful to oversample
|  even more? I mean, two multipaths that are separated by less than half of
|  the chip duration would appear as a single multipath after RRC filtering.
|  Or am I wrong in my reasoning?

You are correct -- multipaths which are separated by less than half a 
chip interval in time will not be individually resolvable.  However, 
oversampling by greater than 2X still can make sense -- it increases the 
accuracy of the matched filter on transmit/receive, and can increase the 
resolution of peak locations for individual multipaths or combined 
multipaths which are being tracked by the rake elements.

   -- Tim Olson
> >You are correct -- multipaths which are separated by less than half a >chip interval in time will not be individually resolvable. However, >oversampling by greater than 2X still can make sense -- it increases the
>accuracy of the matched filter on transmit/receive, and can increase the
>resolution of peak locations for individual multipaths or combined >multipaths which are being tracked by the rake elements. > > -- Tim Olson >
I see. Since you mention increased accuracy, are you assuming RRC filtering in the digital domain? Could you point me to some literature that discusses these matters? It would be interesting to see how the oversampling factor affects performance. I would very much like to know which oversampling factor is likely to be used in practice. /Fred
In article <rfudnUFUnZP3iHHZnZ2dnUVZ_tKdnZ2d@giganews.com>,
 "HSDPA-boy" <fearsome.green@gmail.com> wrote:

|  I see. Since you mention increased accuracy, are you assuming RRC
|  filtering in the digital domain?

Yes.

  
|  Could you point me to some literature that discusses these matters? It
|  would be interesting to see how the oversampling factor affects
|  performance. I would very much like to know which oversampling factor is
|  likely to be used in practice.

2-4X is typical; higher levels of oversampling increase cost (MIPS) vs 
benefit.  If you do a Google search for rake finger oversampling, you'll 
have plenty of papers to read.

   -- Tim Olson
>In article <rfudnUFUnZP3iHHZnZ2dnUVZ_tKdnZ2d@giganews.com>, > "HSDPA-boy" <fearsome.green@gmail.com> wrote: > >| I see. Since you mention increased accuracy, are you assuming RRC >| filtering in the digital domain? > >Yes. > > >| Could you point me to some literature that discusses these matters?
It
>| would be interesting to see how the oversampling factor affects >| performance. I would very much like to know which oversampling factor
is
>| likely to be used in practice. > >2-4X is typical; higher levels of oversampling increase cost (MIPS) vs >benefit. If you do a Google search for rake finger oversampling, you'll
>have plenty of papers to read. > > -- Tim Olson >
Hi gain of the rake fingers is reduced by miss alignment of sampling points and nulls/maximum of the filter. This is reduced if sampling rate is increased. For example gain reduction due to misalignment is -0.91 dB at 4X than -3.9 dB at 2X. Correlation in fingers run at chip rate only. You will find this clearly explained in section 3.4.3 of the book WCDMA requirements and practical design edited by R. Tanner and J. Woodard. --Raghs
>>The use of an oversampled signal when correlating especially >>confuses me
Hello, ideally, if the signals are perfectly aligned and filtered, it does not matter at what rate I do the correlation. This is the case in a typical link level simulation, unless one intentionally uses blocks that delay the signal by a fractional number of samples. Transmitted and received signal are always sampled at the same phase. In an actual link, the sampling phase is not the same, and any time offset deteriorates reception (lower numbers from correlation). One solution is to use a higher a sampling rate, and pick the one that aligns the signals best, as explained in previous posts. BTW: The system response is raised-cosine (two RRCs at Tx/Rx), and together they meet the Nyquist pulse shaping criterion. It means I can sample back down to rate 1 without losing information, *if* the timing (=sampling phase) is accurate. Cheers Markus