How much time does it take to demodulate one OFDM symbol? In 802.11a standard the Symbol Interval Time is 4us. Does this mean that each symbol is demodulated within this period or there is pipelining involved? I understand that the answer may depend on the hardware platform, but I would like to know the range (us/ms) in the fastest implementation. Moreover, a breakdown of time needed at various steps would be nice.
2
-
3\$\begingroup\$ You seem to be asking about latency (rather than throughput), but to get a meaningful answer, you need to specify what you consider the "start" and "end" of the specific process you're interested in. Certainly, pipelining is involved, and some of the pipelines associated with filters and channel equalizers can be fairly long. Also, once a "raw" symbol has been decoded, additional stages associated with error correction are used before you have the actual payload. Please clarify the specific issue you're trying to address here. \$\endgroup\$– Dave TweedCommented Jun 15, 2017 at 1:05
-
\$\begingroup\$ Thanks a lot for the reply. You are right, I am asking about the latency of demodulating one symbol. The 'start' in my process would be the end of the symbol period. And the 'end' would be the decoding of the raw symbol, before error correction. \$\endgroup\$– SiamCommented Jun 15, 2017 at 21:49
Add a comment
|
2 Answers
\$\begingroup\$
\$\endgroup\$
1
Once you have the various I+Q samples, you need to execute an FFT. Plus other "cleanup" steps.
The Wikipedia page states 20MegaSamples/second for 4 us, thus 80 samples. Round up to 128 samples.
If I recall rightly, the computational load for FFT is 4 * N * log2N, thus 4 * 128 * 7 ~~ 3,500 operations. At 1 nanosecond per multiple/accum/butterfly, the time is 3.5uS.
answered Jun 15, 2017 at 4:06
-
\$\begingroup\$ Thanks a lot for the reply. I have few confusions. Is computing the FFT time enough to estimate the total time? Moreover, is the "1 nanosecond per multiple/accum/butterfly" a practical assumption or just an example? \$\endgroup\$– SiamCommented Jun 15, 2017 at 21:46
\$\begingroup\$
\$\endgroup\$
1
There is a lower limit to the time it takes, which is the 4uS it takes to collect all of the data. Obviously you can't start handling the symbol until all the data is in.
There is an upper limit to the time it is allowed to take in a conforming implementation. This is the turn-around time associated with ACK/NACK response. It will be specified in the physical layer protocol somewhere, but I don't know it off the top of my head, and am not about to go digging. At least you now know what you are looking for if you can access a copy.
In between those limits, any time is permitted, and, given the large number of different implementations using FPGAs, DSPs, and custom hardware that exist out there, you can be assured that most possibilities will have been covered.
A better question would be 'how much time does this particular implementation take'? A more meaningful question would be 'how many operations is needed for a symbol decode'?
-
\$\begingroup\$ Thanks a lot for the reply. I have been trying to find the turn-around time associated with ACK/NACK response as you have mentioned. Not much luck yet, but I will keep digging. \$\endgroup\$– SiamCommented Jun 15, 2017 at 22:30