The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl Design >> RF Design >> Channel Filter vs. Matched Filter for FSK https://designers-guide.org/forum/YaBB.pl?num=1238732525 Message started by aaron_do on Apr 2nd, 2009, 9:22pm |
Title: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 2nd, 2009, 9:22pm Hi all, I have a question about direct-conversion FSK. Basically after down-conversion, I have a channel select filter which eases the linearity requirements of the limiting amplifier stage. My question is does the channel-select filter usually have to double as a matched-filter, or do we normally design an additional matched filter after limiting amplification? My thinking is that having an additional matched filter would give better SNR since the limiting amplifier introduces broadband noise. On the other hand, i've seen some designs which do not seem to have the additional matched filter... thanks, Aaron |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by loose-electron on Apr 5th, 2009, 4:42pm Hmmm.... Havent done a FSK system in many years.... A lot of the architectures out there are probably legacy designs. Consequently I would suggest a "never been here before" approach on this. Develop and model at the system level before jumping into transistors. Give some serious thought to direct down convert followed by a minimal BW limiting filter and then ADC, and put all your signal processing in the DSP. Might as well use a modern approach to an old modulation method. Jerry |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 6th, 2009, 12:36am Thanks for the response, but seems like i'm left with more questions than answers :D what do you mean by Legacy design? If I go for direct conversion followed by a minimum bandwidth limiting filter, I guess I will have problems with DC offset. I'm trying to design a low-power receiver. Is ADC + DSP based filtering/detection going to be more power hungry than a analog filtering/detection? I'm actually totally unfamiliar with DSP based design at this stage in my career (the beginning)... Also I selected wideband FSK as I thought the wide tone separation would make filtering (DC offset) as well as the modulation easier. Is there a good reason why not to go for FSK? I know there is an IEEE 802.15.4 (low power) standard out there which uses OQPSK + DSSS, but it seems kind of overkill for a low power low cost application... thanks, Aaron |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 11th, 2009, 1:03am Hi Pancho, I'm only just starting to read up on the modulation/demodulation part of the system, so some of my questions may seem a bit nonsensical... anyway I've done some more reading and perhaps I can re-phrase... Basically I'm looking at 2-FSK with wide tone separation (I guess you pointed me to your other post to illustrate that OQPSK and MSK have similar constellations, right?). Anyway I want the wide tone separation so that when I do direct down-conversion, I can do DC off-set filtering easily. So after my first mixer, I intend to have a channel-select filter which will ease the IIP3 requirements of the limiting amplifier (which follows). The limiting amplifier will have a high gain (maybe more than 60 dB). So after that, the signal needs to be converted into a baseband signal, right? This can be done by a quadrature detector, however, the quadrature detector needs to have more than around 10 dB SNR at the input (in accordance with the Threshold Effect). My question is, "is this threshold effect SNR determined by the noise bandwidth?". If so, is it beneficial to filter off the noise in between the mark and space with an additional filter prior to quadrature detection... Second question is, "would a DSP based design be more power hungry?" i.e. after downconversion, have an anti-alising filter followed by ADC and then DSP... hope my question makes more sense now. thanks, Aaron |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by pancho_hideboo on Apr 11th, 2009, 1:29am Totally I can't understand meanings of your questions. aaron_do wrote on Apr 2nd, 2009, 9:22pm:
aaron_do wrote on Apr 11th, 2009, 1:03am:
They have completely same constellations. MSK signal also can be generated as OQPSK with Halfsine pulse shaping. http://www.designers-guide.org/Forum/YaBB.pl?num=1195226387 aaron_do wrote on Apr 11th, 2009, 1:03am:
aaron_do wrote on Apr 11th, 2009, 1:03am:
aaron_do wrote on Apr 11th, 2009, 1:03am:
Again, it is almost impossible to realize practical matched filter as analog filter. See paper on ZigBee Transceiver which is zero-IF architecture by ADI in ISSCC-2009. [24.4] "A Highly Integrated Low-Power 2.4GHz Transceiver Using a Direct-Conversion Diversity Receiver in 0.18um CMOS for IEEE802.15.4 WPAN" G.Retz1, H.Shanan1, K.Mulvaney1, S.O'Mahony1, M.Chanca2, P.Crowley3, C.Billon1, K.Khan1, P.Quinlan1 1Analog Devices, Cork, Ireland; 2Analog Devices, Valencia, Spain; 3Analog Devices, Limerick, Ireland "A 0.18um CMOS RF sensor-network transceiver based on the 802.15.4-2.4GHz WPAN standard integrates a radio controller, a direct-conversion diversity receiver and a transmitter based on direct VCO modulation. The 5.85mm2 IC draws 16.8mA from 1.8V in receive mode and 18mA at 3dBm output power. The receiver achieves 9.5dB NF, -96dBm sensitivity and -18dBm IIP3, with 54dB of IR at +/-5MHz offset." |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 11th, 2009, 2:23am Thanks for the reply, Quote:
I can't understand what you mean.[/quote] The SNR at the input of the quadrature detector is affected by the noise bandwidth, and I intend to use FSK with wide tone separation. Suppose the data rate is 100 kB/s and the frequency deviation is 500 kHz giving a modulation index of 5. Now suppose the transmission BW is 1.2 MHz. In between the upper and lower frequencies of the 2-FSK signal there should be approximately 800 kHz of BW with a negligible amount of signal information but a non-negligible amount of noise. If I filter off this part of the spectrum, will I improve my overall SNR, which is the same SNR required for the threshold effect (about 10 dB)? The reason i'm not sure is because this in-between noise is at a low frequency deviation and hence may not affect the FM detector at all. But i'm not sure whether it comes into play in determining the SNR for the threshold effect. BTW ISSCC 2009 hasn't been uploaded onto IEEE xplore. I saw a ZigBee design in ISSCC 2008 but it was a low-IF architecture... thanks a lot, Aaron |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by pancho_hideboo on Apr 11th, 2009, 2:33am Maybe you don't understand many isuues about communication theory correctly. aaron_do wrote on Apr 11th, 2009, 2:23am:
Here you have to vary elimination frequency as far as you want to put filter before I/Q-mixer. Such variable band elimination filter is very difficult to realize. Even though you can eliminate noise around center between two frequencies(mark and space) before I/Q-mixer, I/Q-mixer generate noise or offset around DC. aaron_do wrote on Apr 11th, 2009, 2:23am:
|
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 11th, 2009, 2:34am Just saw your last edit...The power consumption seems to be about 3 times higher than the design by Atmel (JSSC, can't remember when but also a ZigBee transceiver). Is there any significant advantage it provides? thanks, Aaron |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 11th, 2009, 2:43am Quote:
Clearly you can see the flaw in my logic. I would be grateful if you could point it out... thanks, Aaron OK...i think i figured it out...kinda |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by pancho_hideboo on Apr 11th, 2009, 2:53am aaron_do wrote on Apr 11th, 2009, 2:34am:
In Atmel transceiver, low IF architecture is used. Remarkable feature of RX is that they use polyphase filter for RF signal path instead of LO signal path. Here complex image rejection filter which is also part of channel selection filter is used. I don't know what design of Atmel you refer, but ADI's design in ISSCC2009 is far low-power than Atmel I know. |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 11th, 2009, 4:17am Quote:
Sorry I remembered wrongly. The Atmel design used 14.7 mA in receive mode (I forgot about the PLL + Digital) and 15.7 mA in receive mode. Still comparable to the paper you mentioned, but i get your point. cheers, Aaron |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by pancho_hideboo on Apr 11th, 2009, 4:19am aaron_do wrote on Apr 11th, 2009, 4:17am:
ADI's design in ISSCC-2009 include much baseband than Atmel. |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 11th, 2009, 4:28am I see. I'll have a look when it is uploaded. Thanks. |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 12th, 2009, 11:30pm Just saw your recent edit... Quote:
I was thinking of going for a direct-conversion FSK. So I can filter the spectrum in between the mark and space by using a bandpass filter centered at a frequency equal to the desired frequency deviation...i.e. the filter would have a center frequency of 500 kHz and BW of 200 kHz. The filtering is done at zero-IF (after the IQ mixers) so there shouldn't be any need for a variable band filter. Does that sound right? BTW, after the limiting amplifier, the quality of the signal is determined by the phase error right? Is there a way to simulate this phase error including the effects of thermal noise? I'm guessing that i need to run a PSS and Pnoise on the limiting amplifier and the rms phase error would be some kind of integration of the phase noise...references would be nice thanks, Aaron |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by pancho_hideboo on Apr 13th, 2009, 5:53pm aaron_do wrote on Apr 12th, 2009, 11:30pm:
aaron_do wrote on Apr 11th, 2009, 1:03am:
You mean frequency discriminator using I and Q signals by "quadrature detection" ? Here I/Q-Mixer exists before "quadrature detection". aaron_do wrote on Apr 12th, 2009, 11:30pm:
aaron_do wrote on Apr 12th, 2009, 11:30pm:
|
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 13th, 2009, 8:29pm Quote:
My mistake...I haven't really got everything figured out yet. I originally believed I could do a direct conversion (I and Q) followed by filtering, and then limiting, and finally quadrature detection. So the filtering would of course be at a fixed IF (zero). I do understand the difficulty in channel filtering at RF...My primary concern was the filtering since I understand I need to have an SNR of around 10dB prior to limiting amplifier, in order to avoid the threshold effect. So my main question was should I filter the spectrum in between the mark and space prior to limiting in order to improve the SNR. I'm really actually trying to figure out what noise BW I should use in order to determine the required NF of the front end. I've concluded that the noise BW in question is just 4*Bmod where Bmod is the bandwidth of the modulating signal (assuming a wide tone separation). After reading more about the mathematics of quadrature detection, however, I guess that quadrature detection should be done at an IF (not zero), right? In which case I would need some other type of detection. Quote:
I'm concerned about when the SNR is around the threshold (10 dB). So does that rule out PSS/Pnoise? thanks, Aaron |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by pancho_hideboo on Apr 13th, 2009, 8:36pm aaron_do wrote on Apr 13th, 2009, 8:29pm:
If you use low-IF, you have to implement image rejection ability. ADI's design of ISSCC2009 is zero-IF where delta-sigma modulator is used to reduce dc-offset and low-frequency noise. aaron_do wrote on Apr 13th, 2009, 8:29pm:
|
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 14th, 2009, 1:19am Quote:
I'm reading the book, "Digital and Analog Communications Systems", by Leon W. Couch II, and his description of quadrature detection is that a quadrature component is first obtained by shifting the signal by 90 deg. So if, Vin = VLcosωct + Θ(t)] then Vquad = K1VLsinωct + Θ(t) +K2dΘ(t)/dt] I assume ωc is a low-IF and not the RF carrier. After that Vquad and Vin are multiplied together and a signal proportional to dΘ(t)/dt results. Since we are using FSK, this product is proportional to m(t)..the original signal. My thinking is that if you do direct downconversion, then the quadrature components which are produced by the down-conversion mixers do not give the additional component proportional to dΘ(t)/dt. So simply multiplying the I and Q components together will not produce m(t). Quote:
I don't necessarily need to use low-IF. I'm just thinking I can't use a quadrature detector (the way used in the book i just mentioned). thanks for the help, Aaron |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by pancho_hideboo on Apr 14th, 2009, 1:57am Your "Quadrature Detection" is classical Quardature Detection or Delay Detection. So my answer of "Not, correct" was not correct. In Quardature Detection or Delay Detection. it requires low-IF. aaron_do wrote on Apr 14th, 2009, 1:19am:
Use I/Q-detection for FSK using I/Q-Mixer, here you can choose both low and zero IF architecture. Also you can treat M-FSK(M>=3). |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by ACWWong on Apr 28th, 2009, 2:51pm Hi Aaron, Direct conversion FSK systems I know which do not use ADC or quadrature detection would use a variant (digital or analog) on the balanced quadricorrelator for demodulation. It is (i believe) the lowest power technique and has been was used in pagers for years (but may suffer too much implemenation loss when the modulation index is very low or frequency offsets are large). I seem to recall that the original paper by Gardner is in "Monolithic Phase-locked Loops and Clock Recovery Circuits: Theory and Design" a collection of papers by Razavi. Anyway I have worked on a zero-IF FSK SoC which is in production and was published in ISSCC08 (paper 7.2) and the wireless part of it (including links to demod) in JSSC July 08 (1st paper). Perhaps it will give you ideas of what can be done in very lower power.... cheers aw |
Title: Re: Channel Filter vs. Matched Filter for FSK Post by aaron_do on Apr 29th, 2009, 12:14am Hi Alan, thanks for the tip. I actually read your ISSCC08 paper a while ago and it was very interesting. I wasn't aware that you had published another paper in JSSC though... cheers, Aaron |
The Designer's Guide Community Forum » Powered by YaBB 2.2.2! YaBB © 2000-2008. All Rights Reserved. |