# DC offset removal from tetra signal

Hey I want to remove DC offset from tetra signal captured as inphase and quadrature phase samples(I and Q samples). I and and Q samples are 16 bit signed values. I tried averaging 2000 I and Q samples out of 2040 sample burst and subtracting corresponding averages from I and Q sample values recieved.But it doesn't seem to work well. Average values I'm calculating seems to vary widely from burst to burst..But if I'm getting DC offset value as average ,it should not be varying much from burst to burst right???

These I and Q samples correspond to pi/4 dqpsk modulated signals .when we average suffiecient number of theses samples to get the DC offset present ,I think we should get a less varying value when we perform same averaging procedure in every burst .. I wanted to know if my reasoning is correct??

could you suggest or point me to any techniques to remove DC ofset from unpredictably varying pi/4 dqpsk modulated I and Q samples. My I and Q sample values are signed 16 bit integer values.So I and Q values could be 32768 to -32767. I see 10481 to -11090 values as my I and Q sample when information is present .When I average 2000 message bearing I samples I get -361 in first time ,10 second time ,-778 next,598 an other time. So I kinda see a variation in average from say -800 to + 800. Could that be the DC offset. I think DC offset should not vary much like above

• usually, to remove DC, we use a simple high-pass filter. apply one HPF to your I signal and another to the Q. – robert bristow-johnson Aug 7 '14 at 5:01
• isn't that exactly I'm doing here. Averaging sufficient number of samples (essentially low pass filtering ) and subtracting I and Q averages from I and Q samples(essentially (1- LPF)=HPF)??? – user46418 Aug 7 '14 at 5:14
• It's okay if there is no message signal present . DC offset introduced by recieve hardware circuitry could be easily removed by the above process. In that case I had to average only 200 sapmles and each time I average I got a value in range 100 to 120 for I sample and -70 to -60 for Q sample. Subtracting those values from I and Q samples is working fine and removing DC offset introduced by recieve hardware circuit irregularities. But now I wanna try removing any inherent DC offset present in the message signal – user46418 Aug 7 '14 at 5:16