[Flexradio] I/Q imbalance and calibration
Put some analysis of the variability of the DDS LO filters out on my website. It's a draft still: comments appreciated. http://home.earthlink.net/~w6rmk/sdr1000/index.htm Take home message so far.. At least from the DDS LO side of things, you don't need a huge number of calibration points across the HF band in order to get good image rejection. {of course, this depends on what you think good is...grin} While there can be pretty substantial differences in phase between I and Q sides due to component variations, once you know what the difference is (e.g. by calibrating), the difference doesn't change a lot over the entire range. If someone has any information on the temperature characteristics of the components (in particular, what's the temperature coefficient of the caps), I can roll that into the analysis. Jim Lux, W6RMK ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://www.mail-archive.com/flexradio%40flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] I/Q imbalance and calibration
Jim, Thank you very much for your mathematical treatment of the I/Q balancing matters. That is really the weak point of the otherwise excellent SDR-1000. To me personally, it means so much that as soon as I realised this more than about 3 years ago I refused to transmit before manually correcting the I/Q balance at the frequency. You know, I am mentally sick in this respect long time! Since the early days of SSB when the signal was generated by home made equipment I had for convenience even front panel controls for phase and balance adjustments! Now that I have been experimenting with Active Integrating Quadrature Sampling Detectors (ISD for short and simplicity) I can clearly see a slanting noise floor graph on the display, raising up towards the increasing negative frequencies, i.e. towards the DC-frequencies. When using 96 or 192 kHz sampling there is a narrower or wider hump visible on the display at around 11 kHz. I have been thinking that the hump is due to higher phase and amplitude erros at low IF-frequencies and due to sampling pulse errors at higher RF-frequencies. Just came into my mind, do we actually need any lowpass filters after the DDS and do we need sinus and cosine outputs? Would just accurately controlled quadrature 180° wide clocking pulses do the trick with a lot less of critical filter components? Then there is some more complication in the QSD circuits. Most of them sample the signal in four short 90° narrow slices when it can be handled by two 180° wide pulses. That is the way I'm sampling in my IDS experiments. Please, wise answers to my silly questions! 73, Ahti OH2RZ On 27/08/06, Jim Lux [EMAIL PROTECTED] wrote: Put some analysis of the variability of the DDS LO filters out on my website. It's a draft still: comments appreciated. http://home.earthlink.net/~w6rmk/sdr1000/index.htm Take home message so far.. At least from the DDS LO side of things, you don't need a huge number of calibration points across the HF band in order to get good image rejection. {of course, this depends on what you think good is...grin} While there can be pretty substantial differences in phase between I and Q sides due to component variations, once you know what the difference is (e.g. by calibrating), the difference doesn't change a lot over the entire range. If someone has any information on the temperature characteristics of the components (in particular, what's the temperature coefficient of the caps), I can roll that into the analysis. Jim Lux, W6RMK ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://www.mail-archive.com/flexradio%40flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://www.mail-archive.com/flexradio%40flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com
Re: [Flexradio] I/Q imbalance and calibration
At 10:27 AM 8/27/2006, Ahti Aintila wrote: Jim, Thank you very much for your mathematical treatment of the I/Q balancing matters. Just you wait.. I'm in the middle of writing up a paper on a calibration model for the SDR1000. The DDS filter analysis was just something that was needed in that process. That is really the weak point of the otherwise excellent SDR-1000. More a weakness in the software, not in the hardware, which is a good thing because it means that it might be correctible in software. The one thing I wish the hardware had was a way to run the Tx and Rx simultaneously, so you could loop the Tx modulator into the receiver. I realize that there are some good, practical reasons why it's not done this way. Now that I have been experimenting with Active Integrating Quadrature Sampling Detectors (ISD for short and simplicity) I can clearly see a slanting noise floor graph on the display, raising up towards the increasing negative frequencies, i.e. towards the DC-frequencies. When using 96 or 192 kHz sampling there is a narrower or wider hump visible on the display at around 11 kHz. I have been thinking that the hump is due to higher phase and amplitude erros at low IF-frequencies and due to sampling pulse errors at higher RF-frequencies. I think you're right. I haven't looked at the audio chain yet, but certainly, for the RF, the sampling error is going to get bigger at higher frequencies (at least measured as phase), because they are probably time offsets (propagation delays, etc.) Just came into my mind, do we actually need any lowpass filters after the DDS and do we need sinus and cosine outputs? Would just accurately controlled quadrature 180° wide clocking pulses do the trick with a lot less of critical filter components? Yes, but it turns out that the easiest way to generate them is to generate a sine with a DDS, filter it, and then threshold. At first glance, it seems that you could take the MSB of the NCO output and use that, but that would mean that your timing resolution is limited to the DDS clock (i.e. 5 ns in the SDR1000 case), so spurs would increase. The use of low order bits essentially provide interpolation to finer than DDS clock rate. Then there is some more complication in the QSD circuits. Most of them sample the signal in four short 90° narrow slices when it can be handled by two 180° wide pulses. That is the way I'm sampling in my IDS experiments. I'm not sure it makes a huge difference. You give up 3dB(?) in energy. It's using +,+,-,-,+,+,-,-,... as the LO waveform rather than +,0,-,0,+,0,-,0,... There's some difference in the harmonic content of the output of the mixer too, but I'm not sure what the impact of that is. That's something that someone has to have analyzed before. Jim, W6RMK ___ FlexRadio mailing list FlexRadio@flex-radio.biz http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://www.mail-archive.com/flexradio%40flex-radio.biz/ FlexRadio Homepage: http://www.flex-radio.com