Filterbank (Gyrator ) for Multiband processing
-
- KVRer
- Topic Starter
- 21 posts since 10 Apr, 2005
Hi all,
I'm trying to implement a multiband device ( more then 2 bands ) into an older dsp.
Everytime i bounce into processor limits.
Is there a way to implement one filter with multiple taps. ie. bass,mid,treb_lo,treb_hi ?
Now i'm just filtering 4 times (2 bands, left + right ) with 15 taps per filter.
like this:
bass_out_left=fir(....,left,...); lowpass
bass_out_right=fir(....,right,...); lowpass
treb_out_left=fir(....,left,...); highpass
treb_out_right=fir(....,right,...); highpass
Is there a faster way?
grt.
gtek
I'm trying to implement a multiband device ( more then 2 bands ) into an older dsp.
Everytime i bounce into processor limits.
Is there a way to implement one filter with multiple taps. ie. bass,mid,treb_lo,treb_hi ?
Now i'm just filtering 4 times (2 bands, left + right ) with 15 taps per filter.
like this:
bass_out_left=fir(....,left,...); lowpass
bass_out_right=fir(....,right,...); lowpass
treb_out_left=fir(....,left,...); highpass
treb_out_right=fir(....,right,...); highpass
Is there a faster way?
grt.
gtek
- u-he
- 28065 posts since 8 Aug, 2002 from Berlin
If your FIR coefficients are symmetric (left side of center looks like right side of center), then the phase would be linear, so you could just substract the lowpass from the input signal to get the highpass.
Cheers,
Urs
Cheers,
Urs
-
- KVRer
- Topic Starter
- 21 posts since 10 Apr, 2005
Yes, that's right.
Treb_out_left=delayed input - bass_out_left;
But, the delaying of the input takes as much cpu cycles as doing a FIR filter action isn't it?
grt Evert
Treb_out_left=delayed input - bass_out_left;
But, the delaying of the input takes as much cpu cycles as doing a FIR filter action isn't it?
grt Evert
2Cooks webradio, late 70's, 80's and pre-90's funk disco and soul muzak.
http://www.2cooks.com:8000/listen.pls
for windows mediaplayer
http://www.2cooks.com:8000
http://www.2cooks.com:8000/listen.pls
for windows mediaplayer
http://www.2cooks.com:8000
- u-he
- 28065 posts since 8 Aug, 2002 from Berlin
That depends... if a delay and a substract is less expensive than twice the multiplies, then it's supposedly faster.
Wait... it IS faster, as you can calculate both at once! Just do the substract while you calculate the lowpass. That should be pretty much faster!
Cheers,
Urs
Wait... it IS faster, as you can calculate both at once! Just do the substract while you calculate the lowpass. That should be pretty much faster!
Cheers,
Urs
-
- KVRer
- Topic Starter
- 21 posts since 10 Apr, 2005
Hmmm,
"Nothing" beats the speed of the Visual dsp's library implemented FIR routine, so this will leave the option for me to recompile the fir source into my own lib.
Well a faster dsp is coming soon, 2000 MFLOPS/s instead of the imaginary 145 MFLOPS/s i do now have. So that will do the trick anyway.
By the way, anyone interested in an ADSP-21061 EZ-KIT?
grt.
Gtek
Evert Verduin
"Nothing" beats the speed of the Visual dsp's library implemented FIR routine, so this will leave the option for me to recompile the fir source into my own lib.
Well a faster dsp is coming soon, 2000 MFLOPS/s instead of the imaginary 145 MFLOPS/s i do now have. So that will do the trick anyway.
By the way, anyone interested in an ADSP-21061 EZ-KIT?
grt.
Gtek
Evert Verduin
2Cooks webradio, late 70's, 80's and pre-90's funk disco and soul muzak.
http://www.2cooks.com:8000/listen.pls
for windows mediaplayer
http://www.2cooks.com:8000
http://www.2cooks.com:8000/listen.pls
for windows mediaplayer
http://www.2cooks.com:8000