Re: [Discuss-gnuradio] Computation of RFNoC Blocks in GRC

[Derek Kozel]

Hi Felix,

The USRPs use signed complex 16 bit integers internally in the FPGA.
These are converted (optionally!) to signed compled 32 bit floating
point numbers on the host pc by UHD. The format actually sent over
USB/Ethernet is called the Over The Wire format and the format that the
user code facing is the CPU format.
https://files.ettus.com/manual/structuhd_1_1stream__args__t.html

These conversions are scaling operations, multiplication. Which is why
your multiplication on the FPGA works as you expect it to. Addition on
the FPGA must be scaled as you found, using the know ledge that a 16 bit
signed integer will be converted to/from a floating point number and
normalized to +-1.

You can set the number formats in the USRP Source and Sink blocks, I
forget if the RFNoC blocks also support that in GNU Radio Companion, but
you can certainly do it in the underlying Python and C++.

Cheers,
Derek

On 13/08/2019 14:12, Felix Greiwe wrote:
> Hello together,
>
> I recently created some RFNoC-Blocks using the RFNoC Modtool and an USRP
> x310 from ETTUS for future GRC application. My Simulation in Vivado went
> well and thus I built an FPGA-Image and flashed it to my USRP x310 device.
> In GRC however, I witnessed very strange behaviour considering addition
> operations of my blocks, so I went back to the most simple functioning
> block, the RFnOC gain block from the RFNoC getting started page:
>
> https://kb.ettus.com/Getting_Started_with_RFNoC_Development
> (All Source Files are linked at the bottom of that page, no need to read
> anything to understand my problem)
>
> What that block does, is to receive a signal from GNURadio, split it into
> I and Q Phase and multiply it by a constant value inside the FPGA of the
> USRP. The outgoing signal is then the product of input and constant, for
> example 0.1*5 = 0.5. Important in this case is, that the range of
> result_values are limited to [-1,1] by the fpga.
>
> To test addition instead of multiplication (which works), i edited one
> single line of the noc_block_gain.v file from
>
>  >> wire [31:0] i_mult_gain = i*gain;
>  >> wire [31:0] q_mult_gain = q*gain;
>
> to
>
>  >> wire [31:0] i_mult_gain = i+gain; // in_phase component (real)
>  >> wire [31:0] q_mult_gain = q*gain; // quadrature phase (imag)
>
> where gain is a 16 Bit register which can be adressed through GNURadio
> Companion.
>
> In GRC I created a simple flowgraph which generates a complex cosine with
> I and Q data samples, which then are prossessed by my modified gain block.
> I expected the Q-Phase to behave exactly like before, and get multiplied
> by the value in the gain register, and the I-Data to move upwoards in my
> GRC Time Sink by the value I add to it.
>
> Link to picture of Flowgraph:
> https://ibb.co/s2jSZtK
>
> While the expectations considering the multiplication got fulfilled, in
> the addition part I see no changes in my results. Only when i pump my gain
> up to great numbers (for example 10000), i see a shift upwoards in my
> diagram. My final observation was, that the shift upwoards is close or
> equal to: "gain_value"/(2^15 -1).
>
> Results with "gain" of 5:
> https://ibb.co/7zJq2zx
>
> Results with "gain" of 26214 (~ (2^15 -1)*0.8):
> https://ibb.co/dtD6cJ7 (Put Q-Data to invisible here, to see I-Data better)
>
> At this point I have a few questions, I hope someone can answer:
>
>  - Why do the operations "multiplication" and "addition" differ in
> behaviour and how can I get my desired "normal" behaviour?
>
>  - At which point does GNURadio/RFNoC/theFPGA convert my signals to the
> range of [-1,1]? Is it always doing this with a fixed formula or only
> when the number exceeds the range? Any other information considering
> this?
>
> - Does it interpret my 16 Bit gain vector as a simple integer in the
> multiplication path and as something different,(maybe Q-Format) in the
> addition path? Why does it not simply add the value of my register to my
> signal?
>
> Any help is greatly appreciated, even when only parts get answered or when
> it addresses errors in my thougthprocess. I did my best to point out my
> problem, feel free to ask if you did not understand everything or my
> mediocre english is an obstacle here.
>
> Best regards and thank you reading,
>
> Felix Greiwe
>
>
>
>
>
> _______________________________________________
> Discuss-gnuradio mailing list
> address@hidden
> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio