Encoding the signal with 14 bits

[f1,Fs1] = wavread('bird.wav');

sound(f1, Fs1);

i tried using

D=int14(f1* 8192)

I get error

Undefined function 'int14' for input arguments of type 'double'.

please help

[Walter Roberson]

You should be considering the Fixed Point Toolbox.

Answer 1

[Star Strider]

I cannot find any reference to int14 in the online MATLAB documentation. If you want to truncate to 14 bit precision, there are two ways to do it with f1 being defined as a column vector. In my experience, .wav files are read as a [N x 2] matrix (one column for each stereo channel), so you will have to encode each column of f1 as 14-bit integers as separate operations. For column 1:

D = dec2bin( (f1(:,1) * 8192), 14);

or if D results a binary string longer than 14 bits, and since binary numbers are defined as strings:

D = dec2bin(f1(:,1) * 8192);
D = D(:,1:14);

then if you want to:

D = int16(bin2dec(D));

They will all be stored as 16-bit signed integers, but truncated in precision to 14 bits.

[Walter Roberson]

Pat wants to encode the signal itself in 14 bits per sample, not in 14 bits precision per sample. Perhaps Pat is running on a machine with a 7 bit byte.

[Star Strider]

That's possible, but a 7-bit byte is not something I can emulate on my machine. I have no idea how MATLAB would deal with that situation.

Anyway, in the absence of Fixed Point Toolbox functions (that I do not have access to and so have no experience with), this is the only way I can envision to produce something similar to what Pat wants.

At least it may provide Pat with some ideas to experiment with. It's the best I can do.

[Pat]

Thanks a lot if not possible to encode the signal directly,is it possible to scale the wavelet coefficients in range[-(2^13-1),(2^13-1) ]

i have decomposed the sound sinal by dwt,so i have wavelet coefficients now

please help

[Star Strider]

What do you want with the wavelet coefficients?

Why are you restricting your data to 14 bit resolution?

[Pat]

i get error

Error using ==> dec2bin at 33 D must be a non-negative integer smaller than 2^52.

Error in ==> sample1 at 68 D = dec2bin( (ys(:,1) * 8192), 14);

[Star Strider]

I do not know what ys(:,1) is, so be sure ys(:,1) > 0 and ys(:,1) < 2^52/8192.

[Walter Roberson]

round() the result of ys(:,1)*8192

[Pat]

[f1,Fs1] = wavread('bird.wav');

ys = DWT2(f1, 100, 1)

ys=(ys*8192)%%%COnverting to 14 bits

now if take minimum value i get

min(ys)=-10495000

max(ys)=10965000

why i am gettin these values ,i need to represent as 14 bits only

[Walter Roberson]

What are max() and min() of the results of DWT2 ?

[Pat]

 min(ys) of dwt2

ans =

 -1.2811e+003

>> max(ys) of dwt2

ans =

1.3385e+003

[Walter Roberson]

Sorry, I do not know how to determine what the theoretical maximum and minimum values returned by DWT2 would be, and I do not have that toolbox to test with.

Looking at that data range, it appears to me that possibly you should take the output of DWT2() and just round() it; it appears that would give you a signed integer with 14 bits needed for its representation (i.e., in the range -2048 to +2047). I cannot make any promises that that pattern will hold.

You would then have to decide which binary representation of negative numbers that you wanted to use when you converted the signed value to unsigned representation.

[Pat]

Walter i get in range Q=round(ys); >> min(Q)

ans =

       -1281

>> max(Q)

ans =

        1339

[Walter Roberson]

Yes, that is consistent with what I wrote.

The range of values (-1281 to 1339) is 2620, which needs 14 bits to represent.

You will need to choose how you want to encode the signed integer values into the "14 bits" that are required for your purposes.

[Pat]

I need it as a signed 14 bit

[Pat]

Just like a signed 16 bit ,i need for signed 14 bit

[Pat]

Walter can u provide with an example plz,because i have no idea regarding encoding the signed integer values into the "14 bits"

[Walter Roberson]

MATLAB does not have a 14 bit representation, except by using the Fixed Point Toolbox.

If you want the data stored within a larger variable (such as 16 bits) please indicate the representation you want to use.

[Pat]

Sorry walter can you tell what you mean by "representation you want to use" as stated by you,because i still i dont have any idea regarding this

[Pat]

I want represent data with signed format

[Walter Roberson]

Separate sign? One's compliment? Two's compliment? Extra bits within the larger word needed to store the 14 bits can be anything, or must be 0, or must be 1, or must be "sign extension" ?

[Pat]

Walter i want to represent as Two's complement .

[Walter Roberson]

And the extra bits in the larger word? The smallest word that MATLAB can use to store 14 bits of information is a 16 bit word: how do you want the two extra bits to be treated?

ys = DWT2(f1, 100, 1);
ys = bitand( typecast(int16(ys),'uint16'), 2^14-1 );

[Pat]

Walter as i said i donthave idea on this as i am new to signals ,can u please what are the ways in which the two extra bits to be treated,please

In

ys = bitand( typecast(int16(ys),'uint16'), 2^14-1 ),the last value must be 2^13-1,dont know it is correct ,please tell

[Walter Roberson]

2^14-1 is correct in that code.

The above code zeros the extra two bits. It's as valid an approach as any other.

Really, things depend on what you are going to do with the 14 bit value.

[Pat]

Thanks walter ,is this same approach can be applied for 16bit

[Walter Roberson]

In 16 bit you do not need those manipulations as datatype int16() or uint16() are already the correct size for that purpose.

[Pat]

ok walter thank u