EMRFD Message Archive 12790
Message Date From Subject 12790 2016-05-04 15:58:15 lmeeny Adjusting IQ Phasing Network in Direct Conversion r\Receiver Hello,
I've designed a phasing network using the QuadNet software. There are 5 single pole all-pass filter stages in the I and Q paths respectively. I can calculate the 90 degree phase shift frequency of each stage.
How do I determine at what frequencies to adjust the phasing network stages for maximum sideband suppression vs. frequency?
Any suggestions appreciated.
12796 2016-05-05 21:10:20 AD7ZU Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver Hi EdFirst determine the bandwidth of the signals you wish to optimize?300hz - 3000hz? a 2700hz bandwidth suitable for SSB maybe CW but not other wider bandwidth modesa CW receiver with a 700hz or 1khz passband will not require 5 stages to get to 50db or greater opposite sideband suppression. but if you want 50db or greater across a 6Khz bandwidth 5 stages may be required.next go to the DESIGN tab in QuadNet and set the bandwidth parameters you wish to see.then go to the SUPPRESSION tab which will display a graph of the all pass network response. .. on the left notice the TUNE button which allows adjustment of each section individually while viewing the graph.this is a quick and visual way to optimize the parameters for each section to meet your design criteria.that's a quick and easy way using features in the QuadNet tool.also note that detectors, mixers, diplexers, and LO signals are not going to be perfect even when built with matched hand selected components and a precision layout so to achieve the kind of numbers shown in QuadNet (-65db) some phase and amplitude adjustment will be needed...somewhere after the mixer / detector and before the IQ signals are combined in a receiver. After IQ signals are generated but before combined in a transmitter.EMRFD has a proof in chapter 9 (i almost called it elegant) that only one adjustment point is required for amplitude and only one adjustment is required to correct the phase error anywhere in the IQ path. Chapter 9 also describes an op amp circuit that may be used to correct the amplitude and phase, however its my take after a brief look at the math involved that the phase correction will only be perfect at one frequency? any comment?? ... not the entire baseband. Additionally It will be challenging to accurately measure 50db or greater opposite sideband suppression across the baseband.hope that helps,RandyAD7ZU
12802 2016-05-07 15:17:26 kb1gmx Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver The way you view that is to insure the two all-pass networks maintain the closest to exactly90 degrees difference between I and Q for the whole as mush as possible of your pass band.Keep in mind that is for the whole pass band and along side as well. For SSB make sureyour wider than say 5khz or what you see is near band sputter from adjacent channels.The other part is if your all-pass were perfect then the preamps and the mixers andthe LO I and Q will be the limiting factors.FYI A 5 pole network for the I and Q likely is overkill. The limiting factor is amplitude and alsodistortion presuming stages before and after the all pass were perfect.EMRFD give you the general patter and the math to calculate what level of phase and amplitudeerror will net what level of performance.from building several more than 43db is very hard to hear a difference and more than 50DB ishard to measure the difference.Allison 12804 2016-05-08 10:20:12 lmeeny Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver Randy,
I've followed the steps you so carefully explained, thank you.
What I'm looking for is a technique as to what pots to adjust at what frequencies. I have an I/Q signal source and can apply the signals at whatever frequencies required.
I sent an email to the guy who wrote the Quadnet program asking him if he could recommend an adjustment procedure. I'll forward his reply. Quadnet does provide the null frequencies. What puzzles me is that several of filter section break frequencies are an order of magnitude above the highest null.
12805 2016-05-08 12:42:15 AD7ZU Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver Hi Ed,the important thing is to get a 90 degree phase shift between the sections.i agree with Allison's comment earlier that unless the bandwidth needed is very wide 5 sections is a lot.and that the allpass network error will likely not be the long pole.. there will be differences in the LO IQ signal phase, duty cycle, detectors/mixers, diplexers and whatever else is in front of the all pass network...there are many sources of error.you need only ONE adjustment for phase and ONE adjustment for amplitude in the entire chain.you may end up with far too many variables to ever get accurately aligned if you build this with pots in each allpass sectionat the very least only put pots in one lane.a cleaner approach is to hand select the caps first (all the same value). you can use a bridge circuit / audio generator / AC volt meter or if you have a rlc meter .. what is important is the phase difference so the cap values are less important than matching the values. If the all pass null is moved a few Hz but is 90 degrees from the opposite lane that is better than trying to precisely set the null frequency. ... actually if you have a big pile of 5% resistors just measure them until you find the values needed. The 3 section network in EMRFD R2 Pro is designed to maximize the use of standard value Rs and a single value Cs as I recall??for SSB and CW a 3 or 4 section allpass network should perform well and I doubt you will hear any difference between a well aligned 3 section and a 5 secti 12806 2016-05-08 14:03:12 Bill Carver Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver Have to disagree, Randy. A four section all-pass has FOUR frequency targets, each is adjusted at its designated frequency to produce precisely 90 degrees difference. When all four are set to the design values, then you get an equiripple phase error........where the deviation from quadrature is equally bad in between the adjustment frequencies.
12807 2016-05-10 12:01:07 lmeeny Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver Randy,
Thanks for the detailed reply.
I realize 5 sections is overkill for 200 - 1800 Hz receiver tuning range but this design is a science project :-) I'm attempting to account for all the gain and phase errors in the chain to determine just how well I can do. The mixer is a Tayloe with hand selected integration capacitors matched to with 0.05%. No diplexers or mixer errors to contend with. The gain resistors in the phase shift circuit are center-tapped precision pairs. Over the frequency of interest the differential gain between the I and Q channels is less than 0.1%. As a first cut I adjusted the series resistor of each stage for 90 degrees +/- 1 degrees at each sections 1/2PI*R*C frequency for a 90 degree +/- 0.6 degree phase shift over the tuning range.
If I understand you correctly I must disagree that a single adjustment for phase in the entire chain is adequate. A two pole phase shifter can do no better than a minimum suppression of -10dB or so.
I agree that having 10 knobs to twiddle I may age a bit before finishing!
Thanks for tossing in you two cents worth :-) The more inputs the more I've learned.
If I can find an adjustment procedure for nulling the unwanted sideband, that is which pot at what frequency I'll post again.
12809 2016-05-10 15:09:21 lmeeny Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver Allison,
Thanks for getting me started. I'm working on a science project ... just how much suppression can be achieved by hand selecting and matching components. with 3 all pass filters in each leg I hit about -40 dB last time around. I've preset the adjustment of each stage in the latest incarnation for a 90 +/- 1 degree degree delay at each filters 1/2*PI*R*C frequency.
12811 2016-05-10 15:56:17 bob_ledoux Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver When experimenting with QuadNet, I saw large theoretically possible attenuations, far beyond those that are practical.
For example, I designed a two pair phase delay network with a bandwidth of 300 to 800 Hz. The calculated suppression was 67.6db for the unwanted sideband. But realistically, I might get 47db of suppression. So I was achieving about 20db less.
Part of the loss is measurement. I don't have access to the tools that can measure L, C, or R values to within .1% The second issue is the parts themselves. voltage, current and change in signal strength.
Back in the 1950's, when crystal filters were in their infancy, a number of manufacturers brought out phasing based SSB rigs. These systems often developed a reputation of being "touchy." They required considerable attention to keep the SSB functions running well.
But they were less expensive to produce than the crystal lattice filters of the era, which required considerable effort to generate the proper frequency differences between the filter elements. Back in those days there was a small contingency of hams who built their crystal filters by taking surplus FT-243 cased filters and polishing them to the desired frequency. The technology of the day required similar processes for commercial equipment which was expensive.
The development of crystal ladder filters made equipment less expensive and made it practical for amateurs to design their own filter systems. This of course also depended on the development of precise measurement equipment.
While a filter constructed of L, C, or R elements is limited by the practical measurement and sources of variances in components, the crystal filter is much less subject to these limitations.
With a simple RF oscillator and a frequency counter today's experimenter can easily select off a set of crystals within a few Hertz of each other. So measured variations of .0001% are practical.
These practical limits, consistent with the nature of electronic components, is one reason phasing filter methods are not common at lower frequencies.
The higher Q of crystals also plays a significant role in filter performance.
This is my view as an experimenter. I'm certain that the older heads with professional credentials can offer an alternative view to my reality.
12812 2016-05-10 18:54:18 kb1gmx Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver If you only hit 40db with a 3 section all pass your doing it wrong. ;)I say that as there are already systems out there doing better with 3 stages.The miniR2 I have runs around 43-45db depending on where in the band it is.The limitation was not the all pass which I think can do better but combinedmixer/preamp/hybrid imperfections.FYI the microR2 with only two sections dialed in does 36-37db.EMRFD has all the dirt on it and what is possible with 1 though 4 stages.Allison 12813 2016-05-10 19:12:26 kb1gmx Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver One comment on the fist phasing rigs.I have a HT37, classic phasing rig that I restored. After replacing the really poorwax-paper caps and the very drifty resistors I dialed it in and despite stories andthe use of tubes it is stable. I can power it off and on and its were it was forsideband rejection. Granted even the 2Q4 needed help due to aging resistorsand caps. The difference was modern carbon and metal film resistors, mylarand polyester film caps. The problem was the quality of the parts used notthe basic scheme. Late 50s and early 60s parts were poor but comparison tothose 20 years later and especially now.As to 1% and .1% parts SMT has been a boon for those. But a digital ohmmeteris still a good tool. Those with modern opamps, hard to beat.The thing is when you get to around 50db of suppression you have toremember that the offending signal is 100,000 time weaker! Its a thing withnumbers is bigger may be better but does it need to be and does it sound good.Crystal filters are easy to make but good crystal filter that age well and havegood shape factors with reasonable group delays are not at all easy to make.Most modern radios the opposing sideband rejection is not the filters resultbut the IF DSP doing phasing in software. The crystal filter in those radioskeeps the clutter out of the IF where it can mix in unpleasant ways.Allison 12814 2016-05-11 11:27:07 AD7ZU Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver Hi Ed,sounds like an interesting project. I have built 2 analog versions of a receiver also using the Tayloe detector. One with 2 sections, the other with 3 sections. I do not know the limit of the Tayloe circuit. I used hand selected parts both R's and C's. There will be differences in switching times and symmetry between on and off times in the switch and in the divider flops and a whole host of other subtle differences that accumulate error.. so at some point an error adjustment is required, and if there is an error adjustment required then all that precision isnt necessary..just tweek out the phase and amplitude error! the second version i built i used the circuit from EMRFD to add a small portion of the opposite lane to correct the phase error. My first take on that solution was that it was just wrong.. but after grinding through the arithmetic it is possible to completely null the phase error at one frequency in the passband...by my math. The amplitude and phase error correction interact with this circuit which means it needs some tedious tweeking. my measurement capability is limited but i could get it to where i could not hear the opposite sideband with more than 200uv at the input through 50db gain. In the 2 section design i added phase adjustments to each section and had similar results. again my measurement capability is limited.The point is that phase and amplitude errors are independent and therefore do not need multiple correction points. That is shown in EMRFD.i did those designs without the aid of quadnet... which is a great tool. one other feature in quadnet is the monte carlo which will allow you to see the calculated results when you provide a tolerance for the Rs and Cs.so.. i moved to a digital solution and used quickfilter technology qf1d512 fir and an audio codec. .. working on a transceiver version now that shares the phase shift and filtering between xmit and rx. ...when things get complicated is always wise to add more variables ;)RandyAD7ZU
12816 2016-05-11 16:50:13 Tayloe, Dan (Noki... Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver
In the NC2030, two sections per side were used to cover the CW bandwidth. The clocking to the detector and the switching timing of the detector bus switch itself introduced phase errors so that a perfect 90 degree I-Q separation was not achieved. In addition, I noticed that when the detector caps are made larger in order to enhance out-of-band signal rejection, the roll off on one sideband is a bit different than the roll off on the other side, which caused an additional phase error effect.
Thus, I put trimmer resistors on the two sections on one side to correct for these distortions. Envision the phase roll off on one side as if it were a straight line phase change with frequency (it is not straight, it is just easier to think of it that way). If the second side starts its straight line phase change descent after the first line, we can get a net phase constant difference between the two sides. We are shooting for something close to a 90 degree shift between these the straight line roll offs of the two sides.
If the clocking source and the detector are not at a 90 degree difference, say 88 degrees instead, then the trimmers in the second side allow the straight line difference to be moved from a 90 degree difference to something different (92 degree?) to compensate for the fact that the source is 88 degree apart, not 90. Also, if the detector or the I/Q preamps introduce unequal frequency roll off (a frequency dependent small phase shift error) between I and Q, the slope of the second side can be changed such that it perhaps produces 92 degrees of shift at the low frequency end and perhaps 91.5 at the high frequency end to try to compensate for some of these additional frequency dependent phase changes.
This form phase compensation is a bit different than what has been discussed since it not only can provide a constant phase difference, but is additionally able to vary the phase difference a bit over the frequency range if needed. I.e., the slope of the two lines can be adjusted so they are a bit different. This adjustment capability can also eliminate the need for high precision matched parts (i.e, the use of 10% caps and 5% resistors) since the scheme takes whatever slope the one side provides, and then matches the second side to it via trimmers to get the desired phase difference. This readily allows one section to use a 0.01 uF cap and the next section to use a 0.1 uF or even a 0.22 uF cap in order to keep the resistance low in that section for noise minimization purposes.
Using surplus HP AC voltmeters that I find at hamfests (~$10, why would anyone want an AC only voltmeter?), I have no problem measuring opposite sideband suppression beyond 60 dB. For the frequency of interest that I was using (CW: ~200 to 800 Hz), I can get >50 dB across the audio band with a bit of interactive tweaking. This phase tweaking was also combined with the ability to tweak the gain of one of the two I/Q preamp stages.
The biggest problem I have had lately in these measurements has been extraneous noise from my cell phone (it tends to be doing all kinds of things in the background), from my nearby Wi-Fi router and from the TV on the other side of the lab wall. Temporarily removal of power solves those issues.
It seems pretty easy to trim phasing networks using two sections per side using a signal generator (8640B) and a HP AC voltmeter. I can see that adjusting three sections on a side would be pretty tough.
- Dan, N7VE
12817 2016-05-11 19:00:03 bob_ledoux Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver Dan makes the case for adjustment, and tuning by ear, rather than relying on high precision parts.
The documents for the NC2030 transceiver remain available here:
Pages 104 through 109 of the Assembly and Operation Manual discuss the adjustment for the best unwanted sideband attenuation. Pages following page 108 provide circuit operation principles. I've also found the Austin and Pacifccon presentations to be very informative.
12818 2016-05-11 19:12:07 AD7ZU Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver Dan,very interesting comment:"In addition, I noticed that when the detector caps are made larger in order to enhance out-of-band signal rejection, the roll off on one sideband is a bit different than the roll off on the other side, which caused an additional phase error effect."is this due to the difference in reactance between sampling caps between the upper and lower sideband frequencies?or is this due to other factors in the detector op amp circuit? or internal op amp characteristics?Thanks,RandyAD7ZU
12819 2016-05-11 19:32:43 Tayloe, Dan (Noki... Re: Adjusting IQ Phasing Network in Direct Conversion r\Receiver
I am not sure. I am not as mathematically inclined as others in the group. Ever played tennis or ping-pong? When I first observed this, my first thought was that it might be the electrical equivalent of “back spin”, helping the signal a little bit on one side, hurting it a little bit on the other.
I have seen this effect over and over again. It is very repeatable. An HP AC voltmeter is your friend. I would never notice this “by ear”.
Again, with a signal generator like my old 8640Bs, it is easy to see 60+ dB down on these phasing networks. When you get past 40 dB of opposite sideband attenuation, just crank in another 10, 20, or 30 dB of RF signal as needed. In my case I am looking at a small signal (200 mV?) at the other end of a 90 dB gain chain. In these tests I am not going to overload the front end.
- Dan, N7VE