See CHANGELOG.md for a summary of the changes.

Change-Id: If1b8f7f2010de93e1328524b761dbb216960a60f

Update CI to run on merge requests

See merge request !36

Will Fantom authored Apr 16, 2025 and Kevin Mooney committed Apr 16, 2025

# CI Updates

This adds the rule: `if: $CI_PIPELINE_SOURCE == 'merge_request_event'` to allow the build pipeline to run on merge requests.

The GitLab docs state "The parent project’s `.gitlab-ci.yml` file must be configured to run jobs in merge request pipelines.", and this change intends to implement this requirement.

> Note, that since this is not already in the parent, this will not be a noticeable change for this specific MR, only MRs post the merge of this.

Interface changes to 'armral_ldpc_decode_block'

See merge request !35

4g5g Consultants authored Apr 02, 2025 and Kevin Mooney committed Apr 02, 2025

> * Added argument `n` , it is the actual length of the RateDematcher output, or the ldpc encoder output length.
> * Added argument `len_filler_bits` , filler bits insertion is needed to ensure that the code block segments have a valid length and are a multiple of the lifting size. Filler bits are used to calculate CRC internally.
> * Replaced argument `crc_idx` with `options`.
>
>   It is an OR'd result of the below fields,
>
>   CRC Type:
>   * ARMRAL_LDPC_CRC_NO
>   * ARMRAL_LDPC_CRC_16
>   * ARMRAL_LDPC_CRC_24A
>   * ARMRAL_LDPC_CRC_24B (default)
>
>   CRC Mode:
>   * ARMRAL_LDPC_CRC_EVERY_ITER (default)
>   * ARMRAL_LDPC_CRC_END_ITER
>
>   Filler Bits:
>   * ARMRAL_LDPC_FILLER_BITS_IMPLICIT
>   * ARMRAL_LDPC_FILLER_BITS_EXPLICIT (default)

Implemented a new algorithm for ldpc decoding

See merge request !32

4g5g Consultants authored Feb 28, 2025 and Kevin Mooney committed Feb 28, 2025

> LDPC decoder algorithm implements iterative message-passing layered decoding. Algorithm involves exchanging the belief information among the variable nodes and check nodes.
>
> Input LLRs "l" provide intrinsic information. Extrinsic information "r" generated in check node is passed to variable nodes.
>
> Belief information is updated iteratively in two steps:
>
> 1. Variable-to-Check step: Variable nodes transmit their belief information to the connected check nodes.
> 2. Check-to-Variable step: Check nodes process the received information, update it, and send it back to the connected variable nodes.
>
> The offset min-sum algorithm is chosen for its computational efficiency and ease of implementation while maintaining reliable decoding performance.

**Note:** _This patch removes the parity_check, as the CRC check is performed which is mandatory._

BER rates with the new implementation are shown in the table below,

| n | bg | mod_type | rv | Eb/N0 | ber_before | ber_after |
|---|----|----------|----|-------|------------|-----------|
| 300 | 2 | QPSK | 0 | 1.0 | 0.217548 | 0.019789 |
| 396 | 1 | QPSK | 0 | 1.0 | 0.211515 | 0.027611 |
| 1400 | 2 | QPSK | 0 | 1.0 | 0.279014 | 0.008802 |
| 1848 | 1 | QPSK | 0 | 1.0 | 0.251669 | 0.017381 |
| 4400 | 2 | QPSK | 0 | 1.0 | 0.321063 | 0.002338 |
| 5808 | 1 | QPSK | 0 | 1.0 | 0.253866 | 0.008661 |
| 10400 | 2 | QPSK | 0 | 1.0 | 0.332846 | 0.000411 |
| 13728 | 1 | QPSK | 0 | 1.0 | 0.254120 | 0.004200 |

Release 25.01

See merge request !29

Fix LDPC decoding accuracy

See merge request !26

A poster on our forum observed that the accuracy of our LDPC decoder
was very much worse than that of other implementations.

The cause is that we do all our calculations in int8_t fixed-point,
which quickly saturates. The fix is to do all intermediate
calculations in int16_t fixed-point.

On average this causes a performance regression of between 1.2x and
1.3x across all ArmRAL benchmarks.

Fix links in Release Notes

See merge request !25

Release 24.10

See merge request !24

Co-Authored-By: Radu Salavat <radu.salavat@arm.com>
Co-Authored-By: Rosie Sumpter <rosie.sumpter@arm.com>
Co-Authored-By: Amy Wignall <amy.wignall@arm.com>

Increase FFT coverage

See merge request !23

Remove duplicate benchmarking cases for matmul_ahb

See merge request !22

Updates a check in the bench.py file to ensure we don't add duplicate
cases.

Release 24.07

See merge request !21

Co-Authored-By: Nick Dingle <nick.dingle@arm.com>
Co-Authored-By: Amy Wignall <amy.wignall@arm.com>
Co-Authored-By: Radu Salavat <radu.salavat@arm.com>

Run coverage on every commit

See merge request !20

This will enable us to automatically gather coverage statistics over
time.

Add CMake option to install docs

See merge request !19

Use CMake's GNUInstallDirs

See merge request !18

Release 24.04

See merge request !17

Co-Authored-By: Rosie Sumpter <rosie.sumpter@arm.com>
Co-Authored-By: Ryo Suzuki <ryo.suzuki@arm.com>

Move license to LICENSE.md

See merge request !15

This will allow Gitlab to automatically detect it and link to it from
the front page of the project, which will make it more visible to
users.

Also update the list of third party licenses (currently empty) to
match and update the documentation to point to the new license files.

Add CHANGELOG.md

See merge request !13

This collates all the change information from the Release Notes for
all releases into a single document. Subsequent releases will have
this information added in future.

We also update the Merge Request templates to tell contributors to
update the Changelog.

Revert addition of LICENCE.md symlink

See merge request !12

This reverts merge request !11

Add symlink to license as LICENSE.md

See merge request !11

This enables the link to the licence file from the main Gitlab page for
RAL, which then displays "BSD 3-Clause "New" or "Revised" license".

Add Merge Request templates

See merge request !8