|
Attached two screenshots are the 12m/10m receiver bandpass filter measured in my QMX. Band is configured with my 17/15/12/10m LPF.
Two traces were taken with the identical filter in the same QMX, I think they were taken only a few seconds apart.
The messy one (the first trace) has a nasty dip in the middle of 12m and the passband somehow extends way below what I designed. I think people have seen similar puzzle of slightly different flavor in various versions of the QMX receiver BPF constructions.
Now the pretty trace - the frequency is shifted about 1 MHz lower than what I designed for, but the shape and the bandwidth reasonably match the theoretical response.
So, what did I do? An RF engineering classic finger test. I opened the case while running the hardware diagnostics on my computer, and started touching parts of the board (back side) with finger. I began to see the pretty trace for a moment. The finger moved, and the trace goes back ugly. Repeated that process with smaller fingers to identify the minimum viable touch area that makes the difference just to confirm what I was interfering with.
The answer: the transmit LPF for the selected band (12/10m in this case).
Does this prove anything? It's a little short of proving anything. But strong suspicion goes to unwanted interaction between the residual reactance from the LPF and the BPF.
My transmit LPF is 5th order elliptic LPF. Both elliptic and Chebyshev filters have ripples in the passband, and because of that, the impedance seen from the terminals fluctuate, both the resistance and the reactance parts, even when the other end is perfectly terminated. The difference can be significant in some cases.
Now, my BPF is not same as described in the standard instructions using T50-2. Mine uses two separate cores for the two segments of the inductor, which I believe removed another potential problem: mutual inductance. I encourage those who followed the standard instructions to do their version of this "experiment" to see if at least one of the strange response goes away. I also encourage those who are set up to do simulation to insert an attenuator between the LPF and the BPF, change the filter design to Bessel or Butterworth, intentionally de-tune the filter or introduce some other changes that interferes with the suspected interaction.
|
Just for illustration purpose, this is the Z1, S11 and S21 of the transmit filter used in my QMX, although what is measured in vitro barrack prototype (I can't measure this in vivo). See Z1 plotted on the Smith chart. I didn't place the marker for this discussion, but the impedance trajectory circled around twice within the passband. That is the impedance fluctuation that I suspect is at play in at least some of the strange BPF behavior.
|
I revisited this issue with morning coffee.
The problem of unwanted interactions between the LPF and BPF is most severe where the BPF's center frequency is near the corner frequency of the connected LPF. For example, in the stock QMX configuration, 30/20m BPF's center frequency sits a bit below the corner frequency of 30/20m LPF.
There can be a number of ways to solve this problem:
* insert a buffer amp of a preamp between T/R switch and the BPF (undesirable in terms of receiver linearity in low bands, perhaps preferred on 15m and above)
* insert an attenuator in the same place (ok on low bands, less desirable on high bands)
* design the LPF with less steep cutoff and smaller ripple (not practical to cover two or more bands per filter)
None of these is easy to test in the QMX board. This subject is very suitable for simulation study, but I thought about a quick and dirty way to show something on the bench.
A shunt resistor was placed between RX_IN (where the BPF capacitors are all connected, before the multiplexer) and the ground. The lower the resistor, the closer the total RX BPF response (measured by QMX diagnostic function) to the theoretical response of the BPF superimposed on that of the connected LPF without unwanted interaction.
Attached two screen shots were measured on 20m band, using QMX's stock LPF for that band. The BPF was mine. The first (broad peak) trace was without a shunt resistor. The LPF's impedance is reactive below the resonant frequency of the BPF, so the peak is broadened on the low side. With a shunt resistor of 47 ohm, the interaction is still there but much smaller. You see a classic tuned circuit response, although the peak frequency turned out a bit higher than what I wanted. The bandwidth and the shape of the latter trace is a lot closer to the theoretical response of the BPF than without the shunt R.
You also notice a slight increase in the insertion loss due to the shunt R. That small level of insertion loss is not a big issue in HF, especially low bands. However, 47ohm shunt is not always enough to eliminate this problem. It is a lot better than having none, though. With some re-shuffling of the L/C and the multiplexer circuit, there could be a few possible implementations.
I used 20m for this example because it uses QMX's stock filter, so this problem is not particular to my LPF.
The theory is simple. The LPF's impedance transition from inductive to capacitive in the upper end of the passband, while the resistive component shoots up and down around the same frequency (see the LPF's impedance on Smith chart in my previous post last night). This is what is shifting the BPF's center frequency, broadening the BPF passband on the low side (high side is cut off by the LPF anyway), and creating a dip. The dip is due to the resistive behavior of the LPF impedance. Since the BPF is in series, this is very simple to think in terms of Thevenin equivalent source, where you can simply add the impedance of the LPF and BPF and then form a voltage divider with the load (detector). Inserting a shunt resistor at the junction limits the swing of the LPF Thevenin impedance. As the shunt conductance is increased, the overall loss increases, but the unwanted interaction decreases, better reflecting the designed BPF response.
|
Hi Ryuji,
The Tayloe detector/downconverter is a complex and interesting entity!
What are you taking for the impedance at the input of that circuit?
Dan Tayloe, in private communication with Steve K1RF, confirmed that
it is very high , in effect an open circuit, at the clock frequency,
and very low at frequencies far removed from the clock frequency. The
circuit exhibits an intrinsic BPF quality, but is responsive to
signals at odd harmonics of the clock.
Since QDX and QMX use the Tayloe circuit as a superhet downconverter
to 12 KHZ, its input impedance will be somewhere in between max and
min at the receive frequency. This really complicates the BPF filter
design. A series resonant LC network working into a high impedance
load is ineffective. Parasitic elements play an outsized role.
Further, the BPF and Zin response of the Tayloe circuit moves as the
clock frequency is swept when creating a filter plot.
Taming the poorly known impedance with a resistive pad at the detector
input might be a reasonable thing to do.
JZ
toggle quoted message
Show quoted text
On Tue, Sep 26, 2023 at 10:38?AM Ryuji Suzuki AB1WX <ab1wx@...> wrote:
I revisited this issue with morning coffee.
The problem of unwanted interactions between the LPF and BPF is most severe where the BPF's center frequency is near the corner frequency of the connected LPF. For example, in the stock QMX configuration, 30/20m BPF's center frequency sits a bit below the corner frequency of 30/20m LPF.
There can be a number of ways to solve this problem: * insert a buffer amp of a preamp between T/R switch and the BPF (undesirable in terms of receiver linearity in low bands, perhaps preferred on 15m and above) * insert an attenuator in the same place (ok on low bands, less desirable on high bands) * design the LPF with less steep cutoff and smaller ripple (not practical to cover two or more bands per filter)
None of these is easy to test in the QMX board. This subject is very suitable for simulation study, but I thought about a quick and dirty way to show something on the bench.
A shunt resistor was placed between RX_IN (where the BPF capacitors are all connected, before the multiplexer) and the ground. The lower the resistor, the closer the total RX BPF response (measured by QMX diagnostic function) to the theoretical response of the BPF superimposed on that of the connected LPF without unwanted interaction.
Attached two screen shots were measured on 20m band, using QMX's stock LPF for that band. The BPF was mine. The first (broad peak) trace was without a shunt resistor. The LPF's impedance is reactive below the resonant frequency of the BPF, so the peak is broadened on the low side. With a shunt resistor of 47 ohm, the interaction is still there but much smaller. You see a classic tuned circuit response, although the peak frequency turned out a bit higher than what I wanted. The bandwidth and the shape of the latter trace is a lot closer to the theoretical response of the BPF than without the shunt R.
You also notice a slight increase in the insertion loss due to the shunt R. That small level of insertion loss is not a big issue in HF, especially low bands. However, 47ohm shunt is not always enough to eliminate this problem. It is a lot better than having none, though. With some re-shuffling of the L/C and the multiplexer circuit, there could be a few possible implementations.
I used 20m for this example because it uses QMX's stock filter, so this problem is not particular to my LPF.
The theory is simple. The LPF's impedance transition from inductive to capacitive in the upper end of the passband, while the resistive component shoots up and down around the same frequency (see the LPF's impedance on Smith chart in my previous post last night). This is what is shifting the BPF's center frequency, broadening the BPF passband on the low side (high side is cut off by the LPF anyway), and creating a dip. The dip is due to the resistive behavior of the LPF impedance. Since the BPF is in series, this is very simple to think in terms of Thevenin equivalent source, where you can simply add the impedance of the LPF and BPF and then form a voltage divider with the load (detector). Inserting a shunt resistor at the junction limits the swing of the LPF Thevenin impedance. As the shunt conductance is increased, the overall loss increases, but the unwanted interaction decreases, better reflecting the designed BPF response.
|
The Tayloe detector/downconverter is a complex and interesting entity!
Certainly it is! And remarkably high performance at a low cost and simple implementation!?? ?
Further, the BPF and Zin response of the Tayloe circuit moves as the
clock frequency is swept when creating a filter plot.
I'm not sure this is the case. When you sweep the RF, at each of the 80 points on an RF sweep plot, the receiver frequency is set by setting Clk0/Clk1 to 12kHz below the desired reception frequency, and the Clk2 (as signal generator) is set to 1kHz above the desired reception frequency. It creates a 1kHz audio tone, and a 1-bucket FFT (Goertzel algorithm) is done as a convenient way of extracting an amplitude measurement, converted to dB, and sent to my ASCII-graph engine.?
The reception is always done at the same frequency?as far as the QSD is concerned?so why would the BPF and Zin response of the Tayloe circuit move? The source impedance?of the BPF will do, as would be normal for a filter anyway.
Isn't the QDX/QMX design so deliciously rich and hard to explain perfectly from a theoretical perspective? Between the PA, QSD and SMPS, it provides a wealth of opportunity for investigation and debate. And fortunately, even though the thorough theoretical explanation is tough, the suck-and-see approach to refining circuit values results in a highly performant radio.?
|
Hi JZ,
Thanks for bringing it up. The quadrature mixer side is probably not innocent. Maybe it's not a bad idea to remove the primary hot side of T401 and stick it into NanoVNA to measure. That should tell.
What you (quoting Tayloe) said makes sense since all the outputs of the multiplexer are shunt by capacitors to the ground. The actual out-of-band rejection performance should be better than shown by the QMX RF sweep screen. In that sense, should we worry more about the in-band loss than out of band rejection? Perhaps, but that difference may not matter much.
Now I understand what I am dealing with, I thought to do some classic tweaking and see how acceptable response I can get. It's not perfect, but 60m to 17m are pretty good. 15m is close. It's ok. I can't bring those bands closer not because I didn't try but because I'm working against the LPF's impedance fluctuation and there is no perfect answer. 10 m is perfect. However, 12m is a disaster. This dip at 12m does not move although I can change how sharp the dip is. If I dampen the LPF with my finger, the dip goes away, so it's something to do with the LPF.
I'm getting about 2.5W on 17m on a 7.4V battery. Harmonic is still -60dBc (didn't change the LPF yet).
I think it's about time to take it to the field and make contacts before spending more time on tweak.
|
Greetings Hans!? Yes, the Tayloe circuit is amazing and very robust! It will tolerate all kinds of misapplication and still provides a virtuous performance!
That its BPF response and Zin response slide about the spectrum with its clock frequency is unavoidable. It is simply what happens, determined by theory.
?What that means, practically, is another matter.
?I shared my concerns about this with you a while back, but noted at that time that QDX and the modified QMX were receiving signals very well. I invoked my long standing belief that one should not fix that which is not broken.
My question to Ryuji attempts to assess his understanding of this beautiful but? arcane bit of circuitry. He is very skilled and is a great add to this forum. I am looking forward to his response.
Regards and best?wishes,? JZ
The Tayloe detector/downconverter is a complex and interesting entity!
Certainly it is! And remarkably high performance at a low cost and simple implementation!?? ?
Further, the BPF and Zin response of the Tayloe circuit moves as the
clock frequency is swept when creating a filter plot.
I'm not sure this is the case. When you sweep the RF, at each of the 80 points on an RF sweep plot, the receiver frequency is set by setting Clk0/Clk1 to 12kHz below the desired reception frequency, and the Clk2 (as signal generator) is set to 1kHz above the desired reception frequency. It creates a 1kHz audio tone, and a 1-bucket FFT (Goertzel algorithm) is done as a convenient way of extracting an amplitude measurement, converted to dB, and sent to my ASCII-graph engine.?
The reception is always done at the same frequency?as far as the QSD is concerned?so why would the BPF and Zin response of the Tayloe circuit move? The source impedance?of the BPF will do, as would be normal for a filter anyway.
Isn't the QDX/QMX design so deliciously rich and hard to explain perfectly from a theoretical perspective? Between the PA, QSD and SMPS, it provides a wealth of opportunity for investigation and debate. And fortunately, even though the thorough theoretical explanation is tough, the suck-and-see approach to refining circuit values results in a highly performant radio.?
|
Hans and JZ,
I interpreted the mixer behavior in the context of real life situation where the LO and RX freq are fixed. We have a target signal at the RX freq, and unwanted interference at a certain offset from the RX freq. I think the rejection of the unwanted interference is likely better than shown in the RF sweep plot because the mixer presents a lower impedance at the interferer's frequency depending on how large the offset is.
Deliciously rich it is. The RX sweep plot changes with and without the functioning final transistors in place.
|
JZ, I did not study your previous discussion on this topic, but I think the frequency-dependent input impedance of the multiplexer-based quadrature mixer circuit to be a positive asset rather than a concern. It only helps reduce out-of-band signals depending on how steep the impedance curve is.
|
Yes Ryuji, I agree completely. The only complaining factor is that the series LC? BPF has to deal with that complexity!
JZ
On Tue, Sep 26, 2023, 4:06 PM Ryuji Suzuki AB1WX < ab1wx@...> wrote: JZ, I did not study your previous discussion on this topic, but I think the frequency-dependent input impedance of the multiplexer-based quadrature mixer circuit to be a positive asset rather than a concern. It only helps reduce out-of-band signals depending on how steep the impedance curve is.
|
Indeed. However, I hope you now see that the complexity from the TX LPF side is much worse!
|
Since the input ?impedance response and bandpass properties of the Tayloe circuit change and slide along with the clock frequency, the self-test BPF plots we make with these radios are somewhat illusory. Each data point on the plot is real and valid as it is taken, but the overall network changes a bit as we change the clock frequency. If we are listening at 14 MHz and hoping to see signals at 9 MHz rejected in accordance with the measurement we made at 9 MHz, that expectation is probably wrong. The network that existed as the 9 MHz measurement was made no longer exists when we listen at 12 MHz. How big is the difference between the plot and reality? Are we better or worse off than what the plot suggests? Does it really matter? That is hard to say.?
toggle quoted message
Show quoted text
On Sep 26, 2023, at 4:58 PM, Ryuji Suzuki AB1WX < ab1wx@...> wrote: Indeed. However, I hope you now see that the complexity from the TX LPF side is much worse!
|
Hi John,
I have to disagree with the invalidity of the BPF sweep.? Granted that it might measure total response as the receiver's sensitivity could change, the receiver is being adjusted to the desired reception frequency each time.? This means the clocks to the Tayloe detector are changed so any impedance would follow.? The one point I agree with is that the measured results would not be the same as if the BPF were scanned with a VNA out of the circuit; in this case, that would not matter as the only adjustments available to tweak the response are the L and C values of the BPF.
For the purposes of tweaking the BPF, the included sweeps make total sense to me.? As for modeling the response, that might be a different topic altogether.
73
Evan
AC9TU
|
...when we listen at 14 MHz.
toggle quoted message
Show quoted text
On Sep 26, 2023, at 8:50 PM, John Zbrozek < jdzbrozek@...> wrote: Since the input ?impedance response and bandpass properties of the Tayloe circuit change and slide along with the clock frequency, the self-test BPF plots we make with these radios are somewhat illusory. Each data point on the plot is real and valid as it is taken, but the overall network changes a bit as we change the clock frequency. If we are listening at 14 MHz and hoping to see signals at 9 MHz rejected in accordance with the measurement we made at 9 MHz, that expectation is probably wrong. The network that existed as the 9 MHz measurement was made no longer exists when we listen at 12 MHz. How big is the difference between the plot and reality? Are we better or worse off than what the plot suggests? Does it really matter? That is hard to say.?
JZ On Sep 26, 2023, at 4:58 PM, Ryuji Suzuki AB1WX < ab1wx@...> wrote: Indeed. However, I hope you now see that the complexity from the TX LPF side is much worse!
|
JZ, I already discussed that topic. If what you quoted is true and that’s what I think, the receiver performance is no worse than predicted from the RF sweep trace. How big is the benefit? I don’t know but I’m not too worried about that. Quadrature sampling followed by active roofing filter is fairly robust against overload. One thing I don’t know is whether the receiver gain is appropriate for the high bands. We’ll find that out soon.
I would like to remind you that a major problem in the receiver is the difficulty in making the BPF perform as designed. Contrary to popular belief, Analog and rf engineering should not be about black art or voodoo tweaking but instead should be about understanding what is responsible for the discrepancy between the theory and the measured performance and how to fill that gap.
|
Hi Evan,
The BPF plots do a fine job of assessing receiver sensitivity at any given frequency. That is exactly how they work, and that is very helpful in providing feedback to the builder that he has done his job correctly.
The filter plots do not provide a correct view of the susceptibility of the receiver to unwanted signals distant to the receive frequency, as these plots know nothing about the intrinsic BPF property of the Tayloe circuit.
Further, the plots know nothing of possible interaction between the highly varying Tayloe input impedance and the LPF/BPF chain response. This, precisely because each data point is taken basically right at the clock frequency. The impedance presented by the Tayloe circuit is then the same at each data point.
The plots I have been seeing appear to show receiver BPF responses that are very low Q. Their broad width suggests that they feed into a resistance of several hundred ohms. At the detector side of the input balun it would be 4X higher.?
In reality, at various distances from the clock frequency, the series LC circuit BPF may see anything in the range of a few tens of ohms to a few thousand ohms. But we don't look there. We only look 12 KHz away from the clock, at each and every data point.
Complex stuff...JZ
On Tue, Sep 26, 2023, 9:07 PM Evan Hand < elhandjr@...> wrote: Hi John,
I have to disagree with the invalidity of the BPF sweep.? Granted that it might measure total response as the receiver's sensitivity could change, the receiver is being adjusted to the desired reception frequency each time.? This means the clocks to the Tayloe detector are changed so any impedance would follow.? The one point I agree with is that the measured results would not be the same as if the BPF were scanned with a VNA out of the circuit; in this case, that would not matter as the only adjustments available to tweak the response are the L and C values of the BPF.
For the purposes of tweaking the BPF, the included sweeps make total sense to me.? As for modeling the response, that might be a different topic altogether.
73
Evan
AC9TU
|
JZ, I totally understand your concern, but as far as I'm concerned, that is a non-issue until it is proven to be a problem. It is probably a minor factor in the mess, but it is not the dominant factor as far as I see. You can't get distracted by minor factors and lose sight of the dominant factors... that's probably why a lot of people believe RF is too much voodoo and little theory.
The BPFs are low Q because (1) they are designed that way, and (2) the interaction with the TX LPF's reactance curve makes the BPF even wider passband by partially undoing the series resonance. I thought I explained above. My posts were chronological and they might have taken people to a mini tour to my experimental journey, which I know can be mentally tiring a bit. I can rewrite more clearly after the whole job is done.
As you see in my traces with (1) my finger on the LPF, dampening/detuning the filter operation, and (2) shunt resistor, dampening the LPF impedance swing both show the BPF response close to theoretical one. If you saw a significant deviation there, you would next look into the next suspicious element down on the list, but I'm totally satisfied with those filter responses for a practical transceiver. I only wish we could easily decouple the LPF and BPF on this very tight board space. Without doing something about this LPF-BPF interaction, doing something about the detector impedance will not improve the BPF operation to any appreciable level.
|
ALL:
If I have any question regarding the Zin(fRX - fLO) characteristics, it is more on how we can take advantage of it. If that function is a steep concave, particularly at lower fLO like 3 to 10MHz range, the BPF bank can be made with a single inductor optimized for higher bands and let the mixer reject the unwanted frequency. That would be very nice. It sounds a bit too good to be true to us but it would also be a big strike against someone trying to build a high sample rate wideband receiver.
|
So, I had the 20m sensitivity issues myself, and rather than futzing around with it I had the realization that the RF is passing through a 5th order lpf on its way into the radio, so another series inductor on top of that is really not going to make or break the high frequency rejection on receive. I ripped my L401 out completely and I've had nothing but improvements in performance. If anyone has a SPICE sim or an irl measurement that proves me wrong I'd love to see it but between the TX LPF and the inherent bandpass property of the Tayloe mixer it seems like extra filtering in the receive section just isn't all that necessary.
My station is in EN34, not the most urbanized area in the world but still full of powerful broadcast stations, and my noise floor is clean enough to pick up WSPR beacons in Australia. Again, if anybody has a real world scenario where L401 provides a benefit that's worth the effort of tweaking the windings until it works correctly, I'd love to see it.
|
So, I had the 20m sensitivity issues myself, and rather than futzing around with it I had the realization that the RF is passing through a 5th order lpf on its way into the radio, so another series inductor on top of that is really not going to make or break the high frequency rejection on receive. I ripped my L401 out completely and I've had nothing but improvements in performance. If anyone has a SPICE sim or an irl measurement that proves me wrong I'd love to see it but between the TX LPF and the inherent bandpass property of the Tayloe mixer it seems like extra filtering in the receive section just isn't all that necessary.
I don't agree... the LPF does indeed suffice for rejection of the harmonic content. I believe the BPF in general provides an improvement in performance. The QSD has very high performance but I believe a BPF ahead of it is still a good idea.?
The claimed inherent filtering of the QSD is useful but doesn't save the day totally. Think for example, of a superhet SSB receiver whose front end consists of a diode ring mixer, then there's a crystal ladder IF filter and a product detector. Typical situation. The crystal ladder IF filter provides the SSB sensitivity. The dynamic range and intermodulation performance (IP3) are determined by the characteristics of the diode ring mixer. We don't say that because an IF filter follows it, there is inherent bandpass filtering so we don't need to protect the front end from strong out of band signals. The IF filter provides the desired selectivity but the signals still have to get to it, through the mixer, first.?
I don't see that the QSD is fundamentally different in this regard. The 1:4 MUX switch is configured as a double balanced QSD. The double balance topology gives you good common mode signal rejection (good rejection of AM broadcast signals). The dynamic range and IP performance is very high but it is still fundamentally limited by this mixer; and still protecting it somewhat from out-of-band strong signals will help a little. The performance is determined by how perfect the switches are. They're a lot better (more perfect) than diode switches which is? why the performance is greater than diode ring mixers.?
If the band pass filter is eliminated then you may not notice any performance degradation and perhaps the few dB of rescued insertion loss could improve sensitivity slightly; but at the expense of vulnerability to out of band signals, which you may not notice having their detrimental effect. The same applies to removing front end filtering on a superhet receiver. You do have some other considerations; on a superhet receiver you need to filter out the image frequency. Well on a hard-switching detector like QSD we need to filter out reception at odd harmonics of the LO (the LPF can do this, as you said). You could remove the front end filtering of an SSB receiver and think everything was OK too. But still the designers wouldn't recommend it.?
|
Ryuji Suzuki AB1WX