You are not logged in.
Pages: 1
Hello,
I'm trying to estimate channel matrices using the example code,
warplab_mimo_2x2_example_ChannelEstim.m, and have some
questions:
1. I want to estimate the channel continuously so that the temporal
variations can be observed. In this case, it seems that the TxLength
should be minimized, but at the same, it should be sufficiently long
to capture the channel precisely. My question is how I compute an
appropriate length for it to maximize the channel estimation frequency?
2. There are four parameters effecting the received signal strength.
Could you please give me some ideas how these should be set for
the channel estimation? There should be some good ranges that
AGC gives fine granuarity.
Thank you so much in advance!
Tae Hyun
Offline
1) The current version of WARPLab cannot stream samples continuously. It buffers a maximum of 2^14 received samples at 40MHz (i.e. in real-time), then transfers them to MATLAB much more slowly. It would be straightforward to capture more samples by reducing the effective sampling rate (by adding a decimation filter in the FPGA between the ADC inputs and WARPLab buffers), but this would require modifying the WARPLab Sysgen core.
2) See Melissa's previous post for details on gains; that thread covers lots of other useful WARPLab details too.
Offline
Does it mean that channel can't be estimated in a micro-second scale in WARPLab?
It sounds like I have some chances only if I use other reference design and modify sysgen core for
them to forward the channel estimation results to PC. Any suggestions for channel variation tracking?
Tae Hyun
Offline
In WARPLab the received signal is sampled at 40MHz and samples are stored in a buffer of length equal to 2^14. This means that you can store a signal of duration equal to 2^14/40e6 seconds and a new sample is stored every 1/40e6 seconds. Once the Rx buffer is full you have to read it from MATLAB and this read operation can take up to 700ms.
As Patrick suggested, you can capture samples for a larger time window by reducing the effective sampling rate (adding decimation filter, you will have to modify the WARPLab sysgen core). The granularity can be changed (to micro seconds, for example) by reducing the effective sampling rate from 40MHZ to the sampling rate or granularity your measurement requires. By doing this you will be able to store 2^14 samples at a rate slower than 40MHz hence the time window during which you store received samples increases.
Offline
livekth wrote:
Does it mean that channel can't be estimated in a micro-second scale in WARPLab?
Just the opposite- WARPLab captures samples every 25nsec, so variations on µsec timescales are definitely observable. The current implementation limits the total length of the capture to 2^14 samples, so observing variations longer than 409µsec (2^14 samples * 25nsec/sample) isn't possible.
Offline
Thank you so much.
Tae Hyun
Offline
One more question on the channel estimation.
Now I want to estimate the links from both antennas in 2x2 MIMO setup.
In the channel estimation example, each antenna is sequentially turned on and off.
Apparently, this approach allows some interval between the estimations of each
vector channel:
estiamte vector channel 1 => download (huge delay) => estimate vector channel 2
(TX radio 2 to RX radio 2 & 3) (TX radio 3 to RX radio 2 & 3)
I'm thinking about estimating two channels at the same time. One possible way is
to turn on two TX radios at the same time and send the stream of symbols like:
Radio 2: ~~~~~~~________
Radio 3: _________~~~~~~
where '~~~' are signals and '____' are zero-signals and their lengths are identically 'x'.
After downloading symbols from the receiver, first x symbols are used to extract the
vector channel 1 and the next x symbols are processed to get the vector channel 2.
One big assumption here is that two radios at the transmitter are perfectly synchronized,
and this sounds reasonable.
Can I do this for the channel estimation? Is there anything that I'm missing here, which
is critical?
Tae Hyun
Offline
The scheme you suggest should work fine. In fact, this orthogonal-in-time training is exactly how our real-time PHY works in MIMO mode. Multiple radio boards at a single node are perfectly synchronized; they share sampling & radio reference clocks, and their control signals are driven by the same FPGA logic.
Offline
Pages: 1