You are not logged in.
Hello,
I would like to measure the bit error ratio of a communication between two warp nodes.
I thought that I send predefined packets over the wireless link to calculate, at the receiving node,
the number of bits that have flipped.
In case of a bad header can I still access the payload?
Or does the reference design skip the payload if there is a bad header / bad packet?
Is there another possiblity to measure BER?
Thank you very much!
Regards
Offline
The BER mode of the current PHY is fairly limited. It requires that you write the expected bytes to a buffer at the receiver before the wireless packet is received. We're working on a more flexible BER module that will be part of the next reference design, which we plan to post in the next few weeks.
Calculating BER of packets with bad headers is tricky. The PHY uses fields in the header to determine length and full rate modulation rates. If the header has an error, the PHY can't trust any of its fields, so the PHY resets immediately and ignores the payload. Another even trickier problem is accounting for packets which aren't detected at all. If the channel imposes a deep fade, the PHY may not even attempt to receive a packet. These events must be accounted for by comparing the numbers of Tx and Rx packets at the end of a test.
Offline
Thank you very much!
In the current BER implementation: Are bit errors of bad header packets calculated (in the code the register is read and printed) or are they
simply skipped?
If these packets are skipped is there a simple way to change the behavior such that payloads of bad header packets don't get skipped
(if the length of the packet and the modulation rates are fixed). - I guess therfore I would have to alter the fpga implementation?
Thank you very much!
Regards
Offline
The current BER calculation updates with every byte written to the packet buffer- it only omits the two checksums (16-bit in the header, 32-bit for the packet). So if a bad header is received, its 22 bytes (24-bytes - 2-byte checksum) would be included in the running BER measurement. Not ideal, I know (I wasn't kidding about it being limited).
The updated BER block will have an option to omit header bits entirely, so you can measure BER of just the full-rate modulation. This version of the PHY is already in the repository, but the full XPS project is a work-in-progress.
It would be possible to use fixed modulation and packet lengths, but (as you guessed) it would require minor modifications to the PHY Sysgen model itself.
Offline
Is the full XPS project for the BER measurement available now ?
With 64QAM, on a ping test, a corrupted byte in the payload is seen. (MISCOMPARE at offset xx)
Not clear how this packet got past the PHY payload-checksum test ?
Offline
The latest reference design (v14) works with berMac.c. The setup requires two WARP nodes and a PC connected to a common switch. Both nodes receive Ethernet frames; one transmits each frame wirelessly, while the other tracks differences between the Ethernet and received wireless frames.
You can customize the code to filter which Ethernet packets are used for BER testing (it's set to use only frames with an etertype filed of WARPNET_ETHTYPE_ARBTRAFFIC = 0x9292 by default). You'll need to customize the code to print the statistics you're interested in.
Offline
Folks,
I am trying to measure BER with the v14 ref. design.
From what I have understood, we have to provide the packets to be transmitted to the boards.
How is this done?
I am guessing that it is doable with a packet generation code that sends Ethernet packets to the nodes. Is this right?
Also there is a mention of a BER test setup in the following link:
http://warp.rice.edu/trac/wiki/OFDMRefe … hmarks/v14
It talks about a refdes server application and client script. Are these scripts doing the custom packet creation/transmission
to/from the boards? If yes, can you provide these as examples?
Thanks,
Mustafa
Offline
marslan wrote:
From what I have understood, we have to provide the packets to be transmitted to the boards.
How is this done?
I am guessing that it is doable with a packet generation code that sends Ethernet packets to the nodes. Is this right?
You would need to generate Ethernet traffic on a PC attached to the nodes. We've used custom apps written with libpcap in our experiments.
marslan wrote:
It talks about a refdes server application and client script. Are these scripts doing the custom packet creation/transmission
to/from the boards? If yes, can you provide these as examples?
We're still working on the server/client setup described on that page (which is why we haven't included BER testing in the list of reference design applications yet).
Offline
Thanks, this all makes sense.
I actually managed to write an application by using jpcap.
The thing is that when I was incrementally printing the number of bit errors, they happened to be in chunks of 205 bits
i.e. the receiver was printing bit errors = 205, 410, 615, 820 and etc.
Is this a bug or is it something that you guys have also seen?
Offline
Nevermind.
I guess I was using packets of 80 bytes, it could be a complication due to that.
Now with 800 byte packets, BER results seem to be matching the theoretical curves.
I can send a sample plot to anyone who is interested.
Cheers,
Mustafa
Offline