AWR1843: Difference between CRLB derived theoretically and implemented in demo code

brickporter

Part Number: AWR1843

Hi:

As mentioned in the thread Compiler/AWR1642: questions about "convertSNRdBToVar" algorithm (function) in AWR1642 SRR released C code,

the CRLB can be derived theoretically as

var(f)>=6/(2pi*2pi)/snr/N/(N*N-1). (Eq. a)

However, in the demo code (lab007) of AWR1843, the CRLB is calculated as

/* CRLB for a frequency estimate */
fVar = (float)invSNRlin * (6.0f / ((2.0f*PI_)*(2.0f*PI_))) * recipsp((n_samples*n_samples - 1)); (Eq. b)

I would like to know why (Eq. a) * N = (Eq. b) .

thanks.

over 4 years ago

0 Aniket Mahajan over 4 years ago

TI__Genius 15046 points

Hi Stan,

I will discuss this with the developer of the lab. Let me get back to you by this Friday.

Regards,

Aayush

0 brickporter over 4 years ago in reply to Aniket Mahajan

Intellectual 335 points

Thanks Aayush.

Look forward to your reply

+1 Aniket Mahajan over 4 years ago in reply to brickporter

TI__Genius 15046 points

Hi Stan,

I have spoken with the developer of the lab. He said that the CRLB formula isn't a formal construct here. As he mentioned in the thread you referred to, this is used as a heuristic to convert the snr to a variance figure.

He told me that when he used the actual formula, the variance came out to be too small to be usable in practice. So he modified some parameters of the CRLB for sinusoid formula to get the optimum performance for the lab.

Regards,

Aayush

Sensors

Sensors forum

AWR1843: Difference between CRLB derived theoretically and implemented in demo code