Fixed point coder log10 implementation

I am using 20*log10(val) to get power in a dB

y = 20*log10(3)

But When i am using matlab coder to convert the in c/c++ code.I am gettig an error

Function 'log10' is not defined for values of  class 'embedded.fi'.      

Can someone help me with alternate way of doing dB calculation.

Thank you

Answer 1

[Life is Wonderful]

qsb = fixed.Quantizer(numerictype(1,32,24));

y0 = quantize(qsb,20*fi(log(double(x)),1,32,24)/fi(log(double(10)),1,32,24)); % log(x)/log(10) for log base 10 i.e. log10

y1 = 20*log10(x);

Answer 2

[Darshan Ramakant Bhat]

The function "log10" is supported for code generation in the latest version (R2020b) of MATLAB. You can find the list of supported functions in the below page :

https://www.mathworks.com/help/referencelist.html?type=function&capability=codegen

[Life is Wonderful]

Thanks - I see, is there a way I can use with R2019b ?

Answer 3

[Life is Wonderful]

I made a test and I don't see the expected result

I am still getting the error following error

??? Function 'log10' is not defined for values of class 'embedded.fi'.

My question - did you verify log10 ?

Below is  the matlab version   
>> ver
-----------------------------------------------------------------------------------------------------
MATLAB Version: 9.9.0.1524771 (R2020b) Update 2
MATLAB License Number: 40523231
Operating System: Microsoft Windows 10 Enterprise Version 10.0 (Build 18363)
Java Version: Java 1.8.0_202-b08 with Oracle Corporation Java HotSpot(TM) 64-Bit Server VM mixed mode
-----------------------------------------------------------------------------------------------------
MATLAB                                                Version 9.9         (R2020b)
Fixed-Point Designer                                  Version 7.1         (R2020b)
HDL Coder                                             Version 3.17        (R2020b)
HDL Verifier                                          Version 6.2         (R2020b)
MATLAB Coder                                          Version 5.1         (R2020b)
Parallel Computing Toolbox                            Version 7.3         (R2020b)

[Darshan Ramakant Bhat]

I tried with the below function and it worked in my MATLAB

function y = myLogFunc(x)

y = 20*log10(x);

end

>> codegen -config:lib myLogFunc -args {3} -report
Code generation successful: View report

Please provide me your sample function along with the script to reproduce the issue.

[Life is Wonderful]

i don't see you generating the Fix point matlab file ? For me - the problem is with coder app generating the fixpoint matlab script

Did you notice your C code still had double for nargin? You have to consider sign, WL and FL for your input and then feed to log10 . Another point for your notice - 20*log10 (x) is NOT fixed point-

Several compiler ask to log function as they won't include the log10 in math.h . For example of Linux based implementation.

Try following sequence

fixptcfg = coder.config('fixpt')
fixptcfg.TestBenchName = 'compute_log_tb';
cfg = coder.config('lib');
codegen -float2fixed fixptcfg -config cfg compute_log -args {3}


============= Step3: Generate Fixed Point Code ==============

### Generating Fixed Point MATLAB Code compute_log_fixpt using Proposed Types
### Generating Fixed Point MATLAB Design Wrapper compute_log_wrapper_fixpt
### Generating Mex file for ' compute_log_wrapper_fixpt '
??? Function 'log10' is not defined for values of class 'embedded.fi'.

Error in ==> compute_log_fixpt Line: 10 Column: 28
Code generation failed: View Error Report
Build failed.
Use help codegen for more information on using this command.

[Darshan Ramakant Bhat]

Thanks for attaching the sample script and explaining the issue clearly. I completely missed the fixed point conversion part earlier.

The error is happening because the function "log10()" cannot take the fixed point inputs. If you do "log10(fi(3))" it will error out.

I found below documentations on fixing such errors :

https://www.mathworks.com/help/fixedpoint/ug/what-to-do-when-a-function-is-not-supported-for-fixed-point-conversion.html

https://www.mathworks.com/help/fixedpoint/ug/manual-fixed-point-conversion-best-practices.html

https://www.mathworks.com/help/fixedpoint/ug/implement-fixed-point-log2-using-lookup-table.html

Based on the above doc I modified the function like below :

function y = compute_log(x)
%#codegen
y = 20*fi(log(double(x)),1,16,0);
end

[Life is Wonderful]

I think it is still wrong

y = 20*fi(log(double(3)),1,16,0)
& 
y = 20*log10(3)

[Life is Wonderful]

May I know what is the difference ?

Why No Fraction length ?

How did you arrive to the conclusion ?

what if y = 20*fi(log(double(x)),1,16,FL) is considerd ?if FL is 16- we get overflow but correct conversion value

Can please share the explanation in case of WL = 8,16,24,32,64- what should be the FL ?

What is range , slope , bias , resolution & precision in that case.

When do we see Overflow and Underflow . Why double is considerd that unsupported in fixed point converter?

I think log (Natural logarithm) is unsupported. We need common logarithm base 10.

How enable logging Mode fipref ? How it help me in verification

Can you please help?Thank you!

[Life is Wonderful]

I found the right implementation -

    qsb  = fixed.Quantizer(numerictype(1,32,24));
    y0  = quantize(qsb,20*fi(log(double(x)),1,32,24)/fi(log(double(10)),1,32,24)); % log(x)/log(10) for log base 10 i.e. log10
    y1 = 20*log10(x);

Question - Matlab fixed point suggestion using fi(value) is not reliable .I don't know why it suggest that results in overflow.

You can try fi(3)

WL =16

FL= 13

Signedness =1

Try , you can see overflow. For me its not clear . Please help me

Thanks