I am trying these things because they are easy, they might work, and
we might get a surprise. Let's please remember that the whole genesis
of RADE is because of a surprise, which came when Google researchers
trained ML to decode Codec2 and it sounded better than our
hand-coded version because it performed its own flavor of harmonic extension.
Thanks
Bruce
On Mon, Sep 21, 2026 at 6:00 PM Greg Maxwell <[email protected]> wrote:
>
> On Tue, Sep 22, 2026 at 12:05 AM glen english LIST <[email protected]>
> wrote:
>>
>> let's be mindful of apples for apples
>> while there might be a "6dB SNR gain", the transmitter backoff for the
>> complex modulation schemes takes some of that away.
>>
>> Typically , hard processed SSB might well generated in a 4.5dB peak to
>> average ratio.
>
>
> I believe RADE has a similar PAPR ratio and the RADE training loop
> specifically constrains the transmitter envelope-- so its optimization
> objective is SNR for a limited amplifier already.
>
>
> _______________________________________________
> Freetel-codec2 mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/freetel-codec2
--
Bruce Perens K6BP
_______________________________________________
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
