EMRFD Message Archive 11086

Message Date From Subject
11086 2015-05-16 19:24:57 Nick Kennedy phasing RX all-pole filter design
I've been trying to learn a bit more about phasing receivers before actually trying something in hardware.  I'm looking at the NC2030, MicroR2 and W6JL web version among others for some examples.

EMRFD has an example in figure 3.59 and  some equations on the next page, plus specific implementations in chapter 9.

I've modeled some of the phase shifting networks in LTspice and played with the transfer function in Excel.

Now I'm wondering where to find specific design equations for these filters based on desired BW and, I suppose, opposite sideband suppression.  Or are they designed empirically?

TU/73-

Nick, WA5BDU
11087 2015-05-16 19:53:08 Dave Re: phasing RX all-pole filter design
Nick:
  There is a program called QUAD.EXE that will design the networks given the number of stages, frequency response.  You can also see the effects of component tolerance.   It is not installed on this laptop, so have to wait until I return.  0.1% resistors are now available for a few cents and 1% .01ufD capacitors are also reasonable.

Dave - WB6DHW


11088 2015-05-17 05:45:49 Fernando Krouwel Re: phasing RX all-pole filter design
Dave and Nick, good morning:

In Texas Instruments website there is a very good software named Filterpro (free download) for active filters design. If I understood correctly, it also allows designing of active phasing networks.

I discovered it by watching a N7VE active filters slideshow design about three years ago.

Unfortunatelly, it does not work with Linux (Wine layer), which is the only OS I use since some time ago. If some of you discover how to put it to work with Linux, please tell me how.

73' s
Fernando - PY2FZU


11089 2015-05-17 06:05:16 kb1gmx Re: phasing RX all-pole filter design
The all pass filter is not a band width filter or sideband suppression filter.

First sideband suppression is the product of I phase signal interacting with Q
phase signals.  So with that hint the phase of signals is a critical parameter.
The second part is the amplitude is also critical.

If you re-read the section in EMRFD to get to high opposing sideband rejection
you require very high degree of phase accuracy and amplitude.  There are graphs
and nearly a chapter on what has to be good for the ultimate sideband rejection.

Now with all that said the better the all pass the higher possible opposing sideband rejection but there are practical limits.  The microR2 is maybe 37DB, good but the 
design was simplified to fit a certain format.  The R2 or MiniR2, sand R2Pro
extend that but you still have to get the RF part also exact.  Having built a few
if the RF part is off the MiniR2 goes from the mid to high 40db range  to under
20.  The math in the books makes it clear, error in any part reduces the possible
sideband rejection.

With that said, the all pass "filter" is not a filter but a phase shifter and the accuracy
and range of frequencies it can operate over are a factor.  But, the RF to base band conversion has to be accomplished with good accuracy as well.  This can be done in software (SDR) and while the computer can give you the exact 90 degree audio 
phase shift the error part to be corrected for is in the RF to baseband conversion
where both amplitude and phase accuracy must be correct.

Now if you want to set bandwidth for the receive channel in those designs 
there is an audio filter LC in the R2 flavors and can also be active.  That 
happens after the sideband selection.

Allison

11090 2015-05-17 06:13:57 kb1gmx Re: phasing RX all-pole filter design
Also a linux user for a long time.

The solution I use for those cases is to run a copy of XP in a sandbox (Vmware or virtualbox).  The fault is not wine but the programs, they are crafted to take advantage
winders hidden functions or the latest and greatest winders with new functions, in some
cases the code is just poorly written.

The alternate solution is to find a suitable substitute, they exist.  Many times
rather than fight I just fire up LibreOffice calc and create a calculator for the 
needed function complete with graphing, it does it well.


Allison


11091 2015-05-17 07:36:26 bob_ledoux Re: phasing RX all-pole filter design
As Allison says, the filter is a phase delay network.  Most designs employ an I/Q input of 90 degrees phase shift and the phase network provides an additional 90 degrees of shift.  The accuracy of the shift is critical for proper attenuation of the unwanted I or Q.  Such accuracy depends on precise values of R and C in the filter pairs.  While precise R values can be purchased precise C values are less common and more expensive.

Tonne's QuadNet software allows for phase delay simulations.  I've found these very educational.  These models well illustrate the presentations on page 9.31 of EMRFD.

In application its "how the filter sounds" that is important.  To that end Dan Tayloe's approach in the NC2030 makes sense.  Place a variable trim resistor in each of the stages and tune the trimmer for best attenuation of the unwanted sideband.

In the NC2030 Dan used a high pass filter to remove frequencies below 300 Hz for both side bands.  His 9 pole low pass filter removed both sidebands above 1Khz.  He is left with both sidebands between 300 and 1000 Hz.  The phase delay network removes the unwanted side band in this range.  Of course that is an over simplification as the filters are not  "brick walls."
11092 2015-05-17 08:10:22 kb1gmx Re: phasing RX all-pole filter design
The microR2 uses the network shown in EMRFD to allow some adjustment/correction of
phase and amplitude before the all pass.  I found it helps "dial it in" but if your not close to start with dialed in is not as good as could be.  What your trying to compensate for is imbalance of RF to the mixers in both amplitude and phase plus any losses in filters after the mixers (there more to terminate the DBMs correctly as well as improve close in overload).

I think it important to review all of the various papers written to see that there real 
things that degrade performance are subtle.

Allison
11096 2015-05-17 08:31:58 Nick Kennedy Re: phasing RX all-pole filter design
Tonne's Quadnet.exe software does look very good for designing and analyzing these networks.


You simply enter you number of poles (stages) and desired BW and it shows R & C values and plots the response.

It doesn't show what's going on behind the curtain, but maybe I don't need to know.  ;^)

It's fairly simple to design a single stage for a specific frequency with 90 degree shift, but how the combinations are designed is not very intuitive.  For example, the stages in the microR2 have 90 degree shift points at 6860 Hz and 568 Hz for the series elements in one chain (I or Q) and 1670 Hz and 138 Hz in the other.  And all this results in good sideband suppression over a range of about 300 to 3000 Hz.

If you put their transfer functions (EMRFD eq 3.27) in a math model (Excel, using complex number formulas), and multiply the series pairs together, you can see that their shifts move around over the passband, but the difference between the two stays close to the optimum (270 degrees, I think).

Octave would probably be good for this.  It's still just reverse engineering though -- it doesn't reveal how those good values were arrived at.

73-

Nick, WA5BDU


11097 2015-05-17 08:41:28 John Levreault Re: phasing RX all-pole filter design
See "Electronic Filter Design Handbook" by Arthur B. Williams. I have the first
edition from 1981, and section 7.5 is titled "Wide-Band 90° Phase-Shift Networks".

John NB1I


11098 2015-05-17 08:58:13 Nick Kennedy Re: phasing RX all-pole filter design
Tonne's Quadnet.exe software does look very good for designing and analyzing these networks.


You simply enter you number of poles (stages) and desired BW and it shows R & C values and plots the response.

It doesn't show what's going on behind the curtain, but maybe I don't need to know.  ;^)

It's fairly simple to design a single stage for a specific frequency with 90 degree shift, but how the combinations are designed is not very intuitive.  For example, the stages in the microR2 have 90 degree shift points at 6860 Hz and 568 Hz for the series elements in one chain (I or Q) and 1670 Hz and 138 Hz in the other.  And all this results in good sideband suppression over a range of about 300 to 3000 Hz.

If you put their transfer functions (EMRFD eq 3.27) in a math model (Excel, using complex number formulas), and multiply the series pairs together, you can see that their shifts move around over the passband, but the difference between the two stays close to the optimum (270 degrees, I think).

Octave would probably be good for this.  It's still just reverse engineering though -- it doesn't reveal how those good values were arrived at.

73-

Nick, WA5BDU


11099 2015-05-17 09:47:09 farhanbox@gmail.c... Re: phasing RX all-pole filter design

just to complicate matters, the low and high pass filters introduce phase delay that is frequency dependent too.


but the good news is the phase delay from these is the same on both channels. but it implies the necessity of symmetry in both channels.


- f

11100 2015-05-17 16:31:53 kb1gmx Re: phasing RX all-pole filter design
>>> doesn't show what's going on behind the curtain, but maybe I don't need to know.  ;^)<<<

Therein lies the problem. the all pass is only one of the 90degree  relationship
systems that makes the phasing system go.  There is a whole lot before it 
that can break it. So for the all pass for every frequency there must be 
a 90 degree phase difference and the bandwidth is where that stops 
being true. When it stops being true you get buckshot or unwanted signals 
that are out of band but heard.

 You have to view it as a system and calculating values for a given network 
means little if you are using 10% parts.   A better calculation is to do it 
repeatedly with known values and then vary them (monte carlo random) 
as if they were 1% parts that can fall anywhere in their specified range and 
see the result.  It is possible to design for minimum error over range.

If you build the bestest all pass in the world but the down converter feeding 
I and Q phase is off by 2 degrees and 2DB to the all pass you still get 
poor results.  The computer guys call that GIGO, garbage in, garbage out.
RF works the same way when relationships(phase and amplitude) 
are important.

Then of course you have to maintain that across the RF tuning range.
Going to Tayloe sampling mixers and digital switching only hides part 
of the problem and has many other subtle issues (and doesn't work at VHF).  

That can be things like one DBM getting more power, or more RF or just plain 
off enough to have different loss. A filter with one inductor with more resistance
can do that too.

So you have to do that for the DBM, IF port low pass filter or termination filters
then the audio (baseband) preamps I and Q then the gain stages that follow 
before the all pass sections.  Once you get to the summer/ subtractor your 
done and the rest is detected result save for additional filters for bandwidth 
and maybe gain to drive phone or speaker.

I've only hit the high points.

Allison
11101 2015-05-17 21:46:34 Will Re: phasing RX all-pole filter design
I also only run Linux.

"Fire up a copy of XP in a sand box", sorry but I have assembled all my
computers for last decade or more and none came with a CD with XP on it.
I sure wasn't going to spend big dollars for something that might be
needed a few hours a year.

"Fire up Libre Office" I use a spreadsheet very infrequently as
relearning how to input math into cells defeats me. Today I tried to do
a few simple resonance calcs. One divided by (2*pi*sqrt(L*C) Basic,
simple but what a job,

Now I know it's a long while (half a century) since I did maths and they
have "New Maths" these days but Pi used to be a Constant (22/7 - approx)
but now Calligra says that Pi is a function! (What off?) Still haven't
worked out how to enter exponents but I have other things to do.

Cheers,'
Will




11102 2015-05-18 08:33:06 AD7ZU Re: phasing RX all-pole filter design
A few simple points that I didn’t see in the chain:

The all pass filter component tolerance are not as important as closely matching the components closely to achieve a 90 degree phase difference between the I and Q lanes.  Example, if the filter is to be constructed using .01 caps for each section then measure and select them, same for the resistors.   The results will be good enough If hand selected parts and a tight symmetric layout are used.

The second point is there is a rigorous math description in EMRFD that many may tend to skip, but the point is that the phase and amplitude errors only need to be corrected at one point in the path where before I and Q signals are combined.

 Another suggestion is to buffer the I and Q baseband detector signals and provide access to those signals. The detector / mixer outputs can be used to drive one of the many SDR applications.    The SDR application makes a decent spectrum analyzer which is a useful instrument to adjust a phasing transmitter or a host of other measurements.

just a few thoughts,

Randy
AD7ZU


11103 2015-05-19 14:35:35 kb1gmx Re: phasing RX all-pole filter design
I keep a copy from old machines that are now defunct so I have the install disks
for every OS from dos 3.1 and win 3.1 through win2K.

Then again I also run VAX/VMS on real hardware.

As to math on a spreadsheet.  I'm pushing acient and I still used and learn.
Its my defense again being obsolete.  ;-)

Allison
11104 2015-05-19 14:48:30 kb1gmx Re: phasing RX all-pole filter design
That is very true as you preserve the relationships.  In real systems if the .01uf is really
.0097 and the rest of them are as well it work just as well.  The resistors can be made up and non precision values with series or parallel values to dial them in.

Yes, on correcting amplitude imbalances and phase as well.   In EMRFD there is 
a good circuit to do that.  In practical use the smaller the correction needed the better
as large corrections are multiplied by the gain difference.  For example twisted wire 
hybrids to get the 90degree LO shift are very repeatable for phase but have to be 
dialed in for amplitude... save for when DBMs are used as they are switching devices 
and amplitude changes can also create phase changes (especially with a weak LO, 
less so with saturating LO) due to the diode thresholds.  That part of the subtle things 
often missed.

My first phasing RX was miniR2 and that set the standard for me as it was easily 
duplicated ( deadbugged one) and worked by the numbers.

The math is important.  Once you understand what being muttered in all that 
its clear that the relationships are preserved the system works.


Allison
11105 2015-05-20 02:15:00 Johan Bodin Re: phasing RX all-pole filter design
Yes, Tonne's Quadnet does a very good job of optimizing and analyzing
I/Q audio phase shifters.

The RC constants of the stages in the I and Q channels should be
selected so that the phase shift difference between the channels is as
close to 90 degrees as possible over the band of interest. I don't know
how Tonne's program finds the sweet spots.

The phase response of a single stage is 180deg - 2 * arctan (2 * Pi *f *
R * C) where R and C are the components at the op-amp's + input. The
resistors at the - input are assumed to be equal. The phase response can
be reversed simply by swapping mentioned R and C.

73
Johan SM6LKM
11106 2015-05-20 10:38:34 Tayloe, Dan (Noki... Re: phasing RX all-pole filter design
>The phase response of a single stage is 180deg - 2 * arctan (2 * Pi *f *
>R * C) where R and C are the components at the op-amp's + input. The
>resistors at the - input are assumed to be equal. The phase response can
>be reversed simply by swapping mentioned R and C.
 
From a practical implementation, swapping the Rs and Cs is simpler from a bias perspective.  With the C to ground and R as the input instead of the C, the stages are DC coupled and the bias for the entire phasing strip can simply be taken from the output of the audio preamplifier stage. 
 
Playing with Quadnet (nice tool!), it seems like a 3 pole could work well for a CW only receiver if the audio was narrowed up to something like 300 to 700 Hz.  That is one less stage to tune or to power.  The resistor values get kind of high, but the values can always be scaled. C1*R1 = C2*R2, where C1 and R1 are the starting values, and R2*C2 are the scaled values.    For example 134K and 0.01uF can be scaled by 10x to get 13.4K and 0.1uf.  R2 went down by 10x, so C2 has to increase by 10x to have the same RC time constant.   If you want to use a 0.47 uf cap, then that is 47x from 0.01 uf, so 134K would get reduced by 47x or 2.85K. 
 
The noise voltage generated by these resistors set how much gain is needed in front of the phasing strip to minimize the receiver sensitivity impact of the phasing strip.  This noise voltage increases with the square root of the resistance.  In other words, a 5K resistor is 100x larger than a 50 ohm resistor (i.e., your antenna input noise), so a 5K resistor is 10x (square root of 100) more voltage noise than the 50 ohm noise of the antenna input.  A 134K resistor is squareroot(134K/50) or ~ 52x the thermal noise of a 50 ohm antenna input.
 
I like to keep these phasing resistors in the 1k to 5k range because this lowers the noise contribution of the resistors and can lead to a more sensitive receiver or alternative, allow the input pre-amp gain to be reduced for the same receiver sensitivity.  A reduction in pre-amp gain before we hit the receiver “brick wall” filtering will improve receiver high signal level capabilities. Less gain implies a higher signal will be needed before overload.  Likewise, I would also lower the two recommended 10K feedback resistors in each stage to something like 2.2K for the same noise reason. The feedback resistors contribute noise just like the phasing resistor.  The only real downside to lowering the resistor size is that it takes a higher drive op-amp to drive these lower resistances.  Better op-amps are rated for a 600 ohm load, and I tend to take advantage of that in the first couple of stages, which on a DC receiver includes the phasing strip.
 
  • Dan, N7VE
11107 2015-05-23 10:35:47 Alberto I2PHD Re: phasing RX all-pole filter design
11108 2015-05-26 13:54:12 Cecil Bayona Re: phasing RX all-pole filter design
I ended buying a copy from Amazon in "very good" condition for $10 including shipping, it just arrived and its brand new never used at all, that is a pretty good buy.

11109 2015-05-26 14:05:52 John Levreault Re: phasing RX all-pole filter design
Enjoy! It's my go-to text for all kinds of filter designs.

John NB1I