fpga/usrp3: Fix DC truncation bias for B2xx by adding rounding to DDC chain

[ryanvolz]

Pull Request Details

Description

Currently, the non-RFNOC usrp3 (i.e. B2xx) DDC code truncates to 18 bits from 24 bits after applying the halfband filters and before applying a scaling and final rounding. This truncation introduces a DC bias that is noticeable with small signal levels (or lots of integration), especially with high decimation rates. Changing the truncation to a rounding removes the DC bias.

This is essentially a forward port of a similar prior fix to the usrp2 DDC chain: https://github.com/EttusResearch/fpga/pull/4. (Note: In order to not repeat the mistake of finding a similar issue in another part of the codebase ~8 years later =P, I checked the RFNOC usrp3's DDC to see what it does. Thankfully, it doesn't have this problem since it uses a wider multiplier and thus avoids the truncation.)

Related Issue

Which devices/areas does this affect?

USRP3-based non-RFNOC devices, i.e. B2xx and B2xxmini

Testing Done

I tested this with a B200mini using GNU Radio. Without the radio connected to anything, I tuned with an LO offset of 1 MHz, gain of 0, and high decimation factor (master clock of 40e6, sample rate of 200e3). Using the stock fpga, a DC bias is present:

Using this modification, the DC bias is gone:

Checklist

• [x] I have read the CONTRIBUTING document.
• [x] My code follows the code style of this project. See CODING.md.
• [ ] I have updated the documentation accordingly.
• [ ] I have added tests to cover my changes, and all previous tests pass.
• [ ] I have checked all compat numbers if they need updating (FPGA compat, MPM compat, noc_shell, specific RFNoC block, ...)

[github-actions[bot]]

CLA Assistant Lite bot All contributors have signed the CLA ✍️ ✅

[ryanvolz]

I have read the CLA Document and I hereby sign the CLA

[ryanvolz]

recheck

[mbr0wn]

@ryanvolz FYI we're reviewing this internally. I think this change is OK.

[mbr0wn]

(Note: In order to not repeat the mistake of finding a similar issue in another part of the codebase ~8 years later =P, I checked the RFNOC usrp3's DDC to see what it does. Thankfully, it doesn't have this problem since it uses a wider multiplier and thus avoids the truncation.)

Oh, and we very much appreciate this ^^^ :+1: