[srsran-users] [ZMQ][TB Decode]

[Andre Puschmann]

Hey Nikos,

On 23/12/21 15:07, Nikolaos Apostolakis wrote:
> I am using srsENB that interacts with srsUE through the zMQ driver. I am 
> trying to emulate a scenario that due to bad conditions srsENB can't 
> decode the TB of the PUSCH, which are sent by the UE. Theoretically, eNB 
> should send NACK messages 3ms after the UL grant has been sent provoking 
> retransmissions from the UE side and consequently yielding 0 throughput.
> 
> To simulate this, I implement a 10ms delay within the 
> /srsRAN/lib/src/phy/phch/sch.c#srsran_ulsch_decode.
> 
> srsENB is connected to the srsEPC where an iperf server is running 
> (everything running within Docker containers on the same host). On UE, I 
> execute iperf tests towards this iperf server. I have appropriately set 
> the routing table of the NS of the UE, so that it sends the traffic 
> towards the tun_srsue interface.
> 
> While expecting near to 0 throughput, I still keep receiving close to 
> 4.5Mbps throughput after several executions. Even with higher > 10ms 
> delay, I still observe considerable throughput.

if I understand you correctly, you are delaying the PHICH feedback to 
10ms. But I also understand that the original PUSCH is still received 
and decoded ok, right? So in this case I would expect the UE to resend 
the PUSCH. But in any case, since the original PUSCH is ok, I wouldn't 
expect zero throughput.

> 
> What I would like to ask if it's my theoretical consideration of the 
> NACK mechanisms is false or there is a technical detail that I am missing.

See above. I don't think it should be zero. Clearly there are more side 
effects of the PHICH not being received correctly by the UE and the 
following allocation of PDCCH/PUSCH, etc. But eventuall the HARQ process 
will be reused by the eNB for a fresh UL TB so that would still get you 
some traffic over the link.

> Is the zMQ driver skipping some of the High-PHY procedures? (reading 
> through the code, I have seen it doesn't).

No, there are no shortcuts being made. In fact the PHY doesn't even know 
the radio layer is using ZMQ for IQ transport.

Thanks
Andre