Re: OFDM digital front end

On Feb 19, 1:00 am, Eric Jacobsen <eric.jacob...@xxxxxxxx> wrote:
On Wed, 18 Feb 2009 01:50:32 -0800 (PST), aitezaz....@xxxxxxxxx wrote:
On Feb 18, 2:47 pm, aitezaz....@xxxxxxxxx wrote:
On Feb 18, 2:21 pm, spop...@xxxxxxxxxxxxxx (Steve Pope) wrote:

<aitezaz....@xxxxxxxxx> wrote:
On Feb 18, 1:41 pm, spop...@xxxxxxxxxxxxxx (Steve Pope) wrote:
Are you talking about within the transmitter, or within the
receiver?
well i'm talking about within the receiver.. i state my problem again
at the transmitter the output of the IFFT block is at 500 kHz. I play
back this data at 500 kHz through the DAC. At the receiver i sample it
at 50MHz. Now, my question is should i downsample it directly to 500
kHz with an integer decimation ratio of 100. Or like in single carrier
system, we downsample it to some higher sample per symbol count i.e.. 2
samples per symbol and after recovering the symbol timing, we come to
1 sample per symbol. In ofdm, is there any need of this?

In OFDM (I should say "in a parallel tone modem") it is common for
the receiver FFT to have 2x or 4x the number of bins as the transmit
IFFT.  (The "number of bins in the transmit IFFT being considered
the nearest power of two that is at least as great as the number of
OFDM tones...er, parallel tones.)

You don't have to do this, but you get less bleeding of
unwanted energy into the bins, and the carrier-offset
correction becomes that much less critical.  It goes a
lot smoother with a higher order FFT in the receiver,
even if in the ideal case it works just the same.

Steve

but how we will decimate at a later stage. like how to get 64 sub-
carriers out of 128 sub-carriers as you described?

and another thing steve. do i have to do this 2X or 4X thing with an
almost perfect clock i.e. 10 ppm or 5 ppm . because as far as baseband
simulations are concerned, perfect upsampling and downsampling results
in almost no performance degradation.
for example in matlab if i go through the following chain
out_data = interp(in_data, 100);
decimate(out_data, 100);
it gives me no performance degradation.

OFDM requires synchronization just like any other digital transmission
scheme, so timing synchronization is required.   You'll probably want
to look into the effects of timing misalignment and how closely your
ultimate receive samples need to be the transmitted locations.
Downsampling will work, but which samples to keep will affect the
timing.   The ability to track drift will depend on your message
length, but even with continuous signals drift can be tracked out with
appropriate synchronization algorithms.

Eric Jacobsen
Minister of Algorithms
Abineau Communicationshttp://www.ericjacobsen.org

Blog:http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php

thanks Eric for your reply. actually the tracking of this timing
offset is not a problem for continuous system. But, i'm simulating a
burst mode system in which i cannot afford feedback loops with long
acquisition time. Besides, i have searched over IEEE xplore about this
and almost every paper say that since this is not much a problem so
you can avoid it by using a very fine clock. I tried to simulate the
OFDM baseband system with a 50ppm clock and the result I found were
very poor. So, do you suggest anything for burst mode system? Like in
continuous mode system, an algorithm is available that tracks the
phase difference between adjacent sub-carriers ... but i think it
would require a lot of samples to converge. What you think I should
use (just reference) for burst mode system.

Thank you very much again.
.

Relevant Pages

  • Re: OFDM digital front end
    ... well i'm talking about within the receiver.. ... at the transmitter the output of the IFFT block is at 500 kHz. ... the receiver FFT to have 2x or 4x the number of bins as the transmit ...
    (comp.dsp)
  • Re: Help with real time system performance estimate
    ... detect changes, encode them and transmit them, preferably via ethernet. ... receiver end of the system needs to receive ... and the corresponding digital outputs being activated needs ... UDP potentially seems better as I can then use multicast to send a ...
    (comp.arch.embedded)
  • Re: M20 unmarked cars blotting out mobile phones.
    ... transmit to, so if they are all occupied it must transmit continuously, ... cancelled out at the receiver in a similar way that transmitted audio ... signals are on the same two-wire line. ... The voice channels could also be combined into a single "high ...
    (uk.legal)
  • Re: The transmission range of 0dbm output power
    ... receiver but in an indoor environment. ... The antenna gain is 2 dbi an omni directional ... I have used path loss equation but someone told me that this ... transmit range of 2 km that it is not supporting at all. ...
    (comp.dsp)
  • Re: irq load balancing
    ... receiver and transmit, ... the lock between cpus. ... There are cases when this is a worthwhile tradeoff, but the network stack is not one of them. ...
    (Linux-Kernel)