[Mp4-tech] Re: [Audio][AAC] RMS conformance for 16 bit AAC LC Decoder

[Ralph Sperschneider]

James Hensey wrote:
> Hello Experts,
>
> Adding to a previous question by Shafi.
> The Conformance Standard ISO/IEC 14496-4, states that the maximum 
> allowed RMS should be calculated using the equation
>
> RMS = 2^(-K+1)/sqrt(12)
> where, K is the required accuracy level.
>
> Using the above equation, for K=16,
> Maximum Allowed RMS = 8.8096 e-6
>
> If the number of output bits per sample is 16, then as per the 
> conformance standard and the conformance software, bits 17 to 24 are 
> zero padded.  
> When bits 17 to 24 are zero padded and assuming that all these 8 bits 
> are in error with respect to the reference output, we get the value of 
> RMS error as
> Calculated RMS= 2^(-16)+ 2^(-17)+ 2^(-18)+ 2^(-19) + 2^(-20)+ 2^(-21)+ 
> 2^(-22)+ 2^(-23)
>  = 30.39e-6
> This calculation assumes that the first 16 bits of the generated 
> output are the same as the reference output.
>  
> The calculated value of RMS is clearly more than the maximum RMS error 
> allowed.
>
> The calculated value of RMS can be less than the maximum allowed RMS 
> only when, for an accuracy level of 16 bits the first 18 bits of every 
> sample are the same as the first 18 bits of the reference output.
>
> Is this an accurate interpretation of the text in the Conformance 
> Standard ?
>
> If yes, then as Shafi identified, RMS criterion cannot be satisfied 
> for an accuracy level of 16 bits when the output bits per sample is 
> also 16 bits.
>
> Regards,
> James
>
> > -----Original Message-----
> > From: mp4-tech-bounces lists.mpegif.org 
> <mailto:mp4-tech-bounces lists.mpegif.org>
> > [mailto:mp4-tech-bounces lists.mpegif.org 
> <mailto:mp4-tech-bounces lists.mpegif.org>] On Behalf Of Shafi
> > Ullah Khan
> > Sent: Monday, March 13, 2024 1:54 PM
> > To: MP4-Tech
> > Subject: [Mp4-tech] RMS conformance for 16 bit AAC LC Decoder
> >
> > I have doubt regarding test vector for RMS conformance of
> > 16-bit AAC LC decoder.
> >
> > Does any AAC LC decoder with 16 bit output able to meet RMS
> > requirements (2^(-15)/sqrt(12) == 8.810e-6 == -101.10 dB) for
> > al00_08.wav and al00_32.wav  reference output?
> >
> > For these two test vectors theoretically RMS cannot be less
> > than 8.821e-6 and 8.817e-6 respectively for 16-bit decoder.
> >
> > shafi
> >

Dear James, Shafi,
the described issue is known: For a decoder with a 16 bit output it 
cannot be guarantied to achieve the conformance criterion for k=16.
Note: The conformance is designed to test the decoder accuracy at the 
decoder output. Any post processing (e.g. subsequent quantization) is 
out of scope.
Hence, you can do the following:
a) You test your decoder output prior to any post-quantization (this 
applies, when your decoder output has a precision of more than 16 bit 
(e.g. 24 bit fixed point or floating point).
b) You claim to be conform for less than k=16. In that case, you 
explicitly have to state k (k=16 is default).
Hope this helps,
Ralph
-- 
Dipl.-Ing. Ralph Sperschneider  | Phone: +49 9131 776 344
Fraunhofer IIS                  | Fax:   +49 9131 776 398
Am Wolfsmantel 33               | mailto:ralph.sperschneider iis.fraunhofer.de
D 91058 Erlangen                | http://www.iis.fraunhofer.de/amm/