Re: [Discuss-gnuradio] Requesting help with HPD block

[Martin Braun]

On 17.03.2015 11:48, Richard Bell wrote:
>  [...]
> My issue might stem from a misunderstanding of the HPD parameters. This
> block seems to have been written for OFDM, but I'm using it for single
> carrier QPSK. If someone wouldn't mind looking over my parameters with
> singe carrier in mind to confirm I've set it up correct, I would be
> grateful. My header is 32 bits long and the payload is 160 bytes or 1280
> bits long. The documentation for this block is not very clear on how the
> parameters relate to the system.

Richard,

the block has some added features to handle OFDM, but it wasn't written 
specifically for it. If you haven't read the unit tests for this code, 
you should definitely go there for some input.
(The reason OFDM was considered as a special case is because this makes 
a CP remover block unnecessary, but that's another story).

> The output of the header port is perfect. It is the exactly what I
> expect. I'm showing 3 headers worth of samples in the sink plot, and we
> see exactly 3 headers with tags on the first sample of each header. In
> the payload length portion of the header, the value 160 exists for the
> payload length. My second question is, when generating the header using
> packet_header_default, is the payload represented in bytes or bits?

Whatever your payload parser generates should be the number of *items* 
read through the HPD. What are bits, bytes? The HPD doesn't know any of 
this. It can only relate to the item size you've given it.

> Now if you look at the payload out sink plot, you will see payloads
> separated by zeros. Each payload portion (the non-zero portion) is the
> exact length as my payload should be, 160 bytes or 1280 bits. But you
> can see zeros are allowed through after the 1280th bit and a second
> payload shows up without a tag. This leads me to believe the HPD block
> is interpreting the payload length to be 4 times larger then I intend it
> to be. What parameters am I mixing up to create this?

Yeah, I can see how that's happening. Your payload is being (probably 
correctly) interpreted as 160 bytes, then the HPD is given the number 
160 as a payload length. However, the HPD needs to know the number of 
*items* per payload. 4 == sizeof(float), which you are using.

The reason the OFDM code has no issues here is because it uses its own 
packet header parser, which returns the number of OFDM symbols. That 
really was the intention of the packet header parser architecture.

Now, how come the packet header parser is reporting 160? Is that 
actually correct? It seems that if your payload is 40 floats, 160 bytes 
seems a lot. You'd be storing 4 bytes per float, when typically, you 
have 1 or 2 bits.

However, as a quick hack, edit the packet_header_default to add a 
scaling factor (or just divide by 4 before sending the packet). That 
should fix it temporarily. A more long-term solution would be to add a 
scaling factor to the packet_header_default for these cases -- maybe you 
want to write and upstream this?

> I've played with the ofdm_tx/rx.grc example and confirmed that it will
> keep the zeros from showing up at the payload out port. This confirms my
> misuse of the block. The hard part is porting the OFDM example
> parameters to a single carrier use. Am I correct in setting symbol to 1
> since there are no OFDM symbols?

Here's what should work:

header length: Whatever you have now
items per symbol:
gi: 0
output format: items

Now make sure the header_parser tells the HPD the number of floats it 
needs to consume per payload.

The value reported by the payload is the same unit as the header length.

M

PS: The reason you see zeros between packets is because the ringbuffers 
get initialized with zeros to start with. Eventually, you might see 
residue from other packets.