>On 10/06/2010 09:23 AM, vrm wrote:
>> Hi,
>> I'm pretty new to DSP and I'm trying to implements simple bass/treble
>> control using RBJ audio filter cookbook.
>>
>> Even if I understand nothing to the math behind the coef calculation. I
did
>> a simple implementation in C using 32 bit float, it's working great !
>>
>> What I want to do is to use blackfin fract16 type to do a fixed point
>> implementation of those filters (I got a BF537 based board).
>>
>> I tried to scale down my coefs for being in the range [-1,+1] but I
achieve
>> to have only crappy sound as result (loss of dynamics and a lot of
>> artifacts).
>>
>> I there something to do in order to get those filter working with fixed
>> point, more than scaling coefs to the [-1,+1] range ?
>
>You cannot scale the coefficients, because that changes the filter 
>frequency response.
>
>You need to either pick a fixed-point representation scheme that 
>includes the coefficient values, or you have to do the coefficient 
>multiplication then shift up (usually by one, because the coefficients 
>shouldn't exceed -2.0 to 2.0), or you have to multiply and add in 
>parallel (i.e. x + 0.9x = 1.9x).
>
Thanks Tim, now it's clear in my mind. I'll try that tomorrow.

Julien

BTW the test code : http://ideone.com/c5o35

On 10/06/2010 09:23 AM, vrm wrote:
> Hi,
> I'm pretty new to DSP and I'm trying to implements simple bass/treble
> control using RBJ audio filter cookbook.
>
> Even if I understand nothing to the math behind the coef calculation. I did
> a simple implementation in C using 32 bit float, it's working great !
>
> What I want to do is to use blackfin fract16 type to do a fixed point
> implementation of those filters (I got a BF537 based board).
>
> I tried to scale down my coefs for being in the range [-1,+1] but I achieve
> to have only crappy sound as result (loss of dynamics and a lot of
> artifacts).
>
> I there something to do in order to get those filter working with fixed
> point, more than scaling coefs to the [-1,+1] range ?

You cannot scale the coefficients, because that changes the filter 
frequency response.

You need to either pick a fixed-point representation scheme that 
includes the coefficient values, or you have to do the coefficient 
multiplication then shift up (usually by one, because the coefficients 
shouldn't exceed -2.0 to 2.0), or you have to multiply and add in 
parallel (i.e. x + 0.9x = 1.9x).

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

Hi,
I'm pretty new to DSP and I'm trying to implements simple bass/treble
control using RBJ audio filter cookbook.

Even if I understand nothing to the math behind the coef calculation. I did
a simple implementation in C using 32 bit float, it's working great !

What I want to do is to use blackfin fract16 type to do a fixed point
implementation of those filters (I got a BF537 based board).

I tried to scale down my coefs for being in the range [-1,+1] but I achieve
to have only crappy sound as result (loss of dynamics and a lot of
artifacts).

I there something to do in order to get those filter working with fixed
point, more than scaling coefs to the [-1,+1] range ?

Thanks !
Julien