|
I just finished sweeping the qdxh Tayloe_mixer input impedance with my nanovna. I will post some screen grabs as a series of posts because the images will be too big for one post.
Tony
AD0VC
toggle quoted message
Show quoted text
Ryuji,
I made some improvements to my Tayloe simulation model, mostly in the
area of clock generation. It had a very positive effect on the impulse
response spectrum of the circuit. It is considerably cleaner now.
I also made a few runs specifically to test the input impedance
properties of the circuit both near and at the clock frequency. I did
that by using a 1 amp current source as the input signal and then
measuring the voltage at the input balun as a function of signal
frequency.
The response was amazing! The input impedance rises very rapidly as
the signal frequency approaches the clock frequency, achieving a value
in the high *thousands* of ohms at the clock frequency. A real circuit
would have never survived this experiment!
This was the result described by Dan Tayloe, and which I had been
hoping to see. I was previously thwarted by having some overlap
between clock phases. Clocks matter!
JZ
On Fri, Oct 20, 2023 at 7:13?PM Ryuji Suzuki AB1WX <ab1wx@...> wrote:
>
> Clocked switching itself is linear so that approach/assumption is valid. What cannot be concluded is the mixer's dynamic range or the degree of effects/benefits of having a BPF since all the real life nonlinearity are omitted from the simple model.
>
>
|
JZ, keep up with good work on the refinement. What is the actual complex impedance profile +/- 100kHz of fLO, where fLO is between 3MHz and 30MHz? Then implement realistic strays?
If the impedance is very high and well behaved, after you implement all the strays realistically, one approach to clean up the whole BPF issue while increasing the overall receiver sensitivity may be to increase the input transformer winding ratio and/or replacing the BPF with a high-pass L-match (series C to the source, shunt L to the load) which will behave like a band-pass anyway. I'm not sure what the actual input impedance will be though, I get an impression that this line may be around 75 or 150 ohm, a bit reactive. For the high bands, where I replaced the BPF with a pi-mach, whatever the load impedance is, it doesn't really matter much because when I tweak the pi-match inductor it's close enough for a receiver work.
|
Ryuji,
I made some improvements to my Tayloe simulation model, mostly in the
area of clock generation. It had a very positive effect on the impulse
response spectrum of the circuit. It is considerably cleaner now.
I also made a few runs specifically to test the input impedance
properties of the circuit both near and at the clock frequency. I did
that by using a 1 amp current source as the input signal and then
measuring the voltage at the input balun as a function of signal
frequency.
The response was amazing! The input impedance rises very rapidly as
the signal frequency approaches the clock frequency, achieving a value
in the high *thousands* of ohms at the clock frequency. A real circuit
would have never survived this experiment!
This was the result described by Dan Tayloe, and which I had been
hoping to see. I was previously thwarted by having some overlap
between clock phases. Clocks matter!
JZ
toggle quoted message
Show quoted text
On Fri, Oct 20, 2023 at 7:13?PM Ryuji Suzuki AB1WX <ab1wx@...> wrote:
Clocked switching itself is linear so that approach/assumption is valid. What cannot be concluded is the mixer's dynamic range or the degree of effects/benefits of having a BPF since all the real life nonlinearity are omitted from the simple model.
|
Clocked switching itself is linear so that approach/assumption is valid. What cannot be concluded is the mixer's dynamic range or the degree of effects/benefits of having a BPF since all the real life nonlinearity are omitted from the simple model.
|
Ryuji,? Yes, it is correct that the spectral roll off of the Dirac approximation must be considered when viewing these results.
Regarding other means of seeing the input impedance of the Tayloe, recall also that any LTSpice swept frequency analysis is based on a linearization of the simulated circuit. The complicated action that results from clocked switching is lost in that and results can be quite bizarre.
JZ
On Fri, Oct 20, 2023, 6:29 PM Ryuji Suzuki AB1WX < ab1wx@...> wrote: I see. So what I suggested earlier, dividing the voltage by the current and plot as a complex number would solve the problem... not? (Or you could take magnitude of Fourier transform of the raw fake impulse and divide, but it is less preferable...)
Does that also mean that the true asymptote on the high frequency side is also different from what's shown in the graph?
|
I see. So what I suggested earlier, dividing the voltage by the current and plot as a complex number would solve the problem... not? (Or you could take magnitude of Fourier transform of the raw fake impulse and divide, but it is less preferable...)
Does that also mean that the true asymptote on the high frequency side is also different from what's shown in the graph?
|
Ryuji,
Remember that my Dirac spike is an approximation.? A true Dirac Delta would be flat over the entire spectrum. The approximation does not possess that property. It's spectral response is part of what you are seeing.
Over the last few months I have played with a number of attempts at modeling the Tayloe circuit. Sadly, no LTSpice model for the mux chip exists. I started. by using BS123 models run at low frequency. That was interesting but it had its limitations. Later I tried the chip-level NMOS models in LTSpice, the ones in which you specify device width and channel length. I had some success with that, too. but still parasitics, and especially clock feedthrough to the input were complicating things. I decided to go with the utterly clean and parasitic free switches to best understand the tracking-filter property of the Tayloe circuit. That clarified many things.
JZ
On Fri, Oct 20, 2023, 5:56 PM Ryuji Suzuki AB1WX < ab1wx@...> wrote: JZ, indeed. I'm deeply in LPF/BPF/amp area and I don't know if I want to pick up another project. Well, mixer is related to BPF issue but I'd rather see all the right questions reported here :-)
So, are you saying that input Z is inductive where f < f_LO, and capacitive otherwise? But then your first trace goes up as f goes below 100k. Isn't that capacitive??
Real transformers have coupling coefficient < 1 and trifilar on a FT50-43 is not leakproof, and that transformer has a lot of magnetizing inductance. So, I would expect the real mixer to have some leakage inductance, especially at higher frequencies. Then there is winding capacitance. So, I would expect the mixer input impedance to be also signal frequency dependent not just relative to the LO frequency. I did note that you made a caveat that parasitics were omitted, but the real simulation needs parasitics...
|
JZ, indeed. I'm deeply in LPF/BPF/amp area and I don't know if I want to pick up another project. Well, mixer is related to BPF issue but I'd rather see all the right questions reported here :-)
So, are you saying that input Z is inductive where f < f_LO, and capacitive otherwise? But then your first trace goes up as f goes below 100k. Isn't that capacitive??
Real transformers have coupling coefficient < 1 and trifilar on a FT50-43 is not leakproof, and that transformer has a lot of magnetizing inductance. So, I would expect the real mixer to have some leakage inductance, especially at higher frequencies. Then there is winding capacitance. So, I would expect the mixer input impedance to be also signal frequency dependent not just relative to the LO frequency. I did note that you made a caveat that parasitics were omitted, but the real simulation needs parasitics...
|
Lastly, the coupling coefficient I used is 1.
On Fri, Oct 20, 2023, 5:22 PM Ryuji Suzuki AB1WX < ab1wx@...> wrote: JZ,
Is n002 an input node with a purely resistive constant impedance to the ground? Why not divide with the input current and plot in the polar or re/im coordinates?
What's the coupling coefficient of the input transformer in your model?
|
Hi Ryuji,
The Tayloe input impedance is complex, with a response similar to a high Q parallel LC circuit, and it moves about with the clock frequency.
I have used several different schemes to illuminate the behavior of the Tayloe input, including direct ratios, injection of wide band noise from a behavioral voltage or current source, and finally the Dirac Delta function approach. The last produces the best results. The others are computationally burdened or otherwise compromised.
If you are interested I can send you my sim deck.
I am delighted to see the elevation of technical discussion that is taking place on this forum!
JZ KJ4A?
On Fri, Oct 20, 2023, 5:22 PM Ryuji Suzuki AB1WX < ab1wx@...> wrote: JZ,
Is n002 an input node with a purely resistive constant impedance to the ground? Why not divide with the input current and plot in the polar or re/im coordinates?
What's the coupling coefficient of the input transformer in your model?
|
JZ,
Is n002 an input node with a purely resistive constant impedance to the ground? Why not divide with the input current and plot in the polar or re/im coordinates?
What's the coupling coefficient of the input transformer in your model?
|
Thanks John. Appreciated.?
Peter
|
Peter,
In my morning I will DM you a .asc file for that simulation. Have fun with it!
The literature regarding the input impedance of the Tayloe circuit is nil it seems, as is that for optimum clock phase waveforms/overlap. These are important issues that are not well considered.? Simulation may help clear that up.
JZ KJ4A?
On Fri, Oct 20, 2023, 3:07 PM Peter Ayearst < ve3poa@...> wrote: John,
Lately I've been thinking about the Tayloe Detector and setting up a LTspice simulation.? Could you post a clearer picture of your schematic?? I'd be most interested in seeing what you have done.
73, Peter ve3poa?
|
John,
Lately I've been thinking about the Tayloe Detector and setting up a LTspice simulation.? Could you post a clearer picture of your schematic?? I'd be most interested in seeing what you have done.
73, Peter ve3poa?
|
I thought it might be useful to post the sim results for detection of
a 1.001 MHz signal vs the 1.000 MHz clock. The Cacc waveforms are
beautiful, representing a 1KHz beat note with I and Q and their
complements present.
Enjoy!, JZ KJ4A
toggle quoted message
Show quoted text
On Fri, Oct 20, 2023 at 12:10?PM John Zbrozek <jdzbrozek@...> wrote:
Stephan, you had asked for some simulation results.
I couldn't resist ;-)
The LTspice simulation below shows some of the fascinating properties
of the Tayloe detector.
The simulation uses ideal voltage controlled switches to perform the
MUX function of the detector. This avoids complications with
parasitics and device performance.The switch model also has an
optional internal resistance. Since the switches change state 'hard'
at defined voltage points, the four phase clock system can simply be
four sine generators staggered by phase. I used 2 X 4 just to keep the
visual clutter down. As shown, the clock is running at 1MHz.
The input signal is a single sharp spike, an approximation to the
Dirac Delta function, which produces a very wide spectrum.
The simulation output is selected as the voltage at the input winding
of the balun. As you might expect, it is also a single sharp spike.
The voltage at that node can be taken as a surrogate for the input
impedance of the detector.
The real treasure of the simulation is the FFT spectrum at that node,
shown. You can clearly see the intrinsic bandpass behavior of the
Tayloe circuit, as well as its receptivity at odd clock harmonics. As
the clock frequency changes, the response curve follows it. As the
Cacc parameter alters the values of the accumulating capacitors, the
bandwidth changes.
JZ KJ4A
On Fri, Oct 20, 2023 at 3:17?AM Alan G4ZFQ <alan4alan@...> wrote:
On 19/10/2023 16:35, Rod Smith via groups.io wrote:
does this link work for you?
<>
Rod,
Yes, thanks.
An impressive project for it's time. Some parts can be improved and
simplified by modern developments.
The designers certainly believed in good filtering, no compromises.
73 Alan G4ZFQ
|
Stephan, you had asked for some simulation results.
I couldn't resist ;-)
The LTspice simulation below shows some of the fascinating properties
of the Tayloe detector.
The simulation uses ideal voltage controlled switches to perform the
MUX function of the detector. This avoids complications with
parasitics and device performance.The switch model also has an
optional internal resistance. Since the switches change state 'hard'
at defined voltage points, the four phase clock system can simply be
four sine generators staggered by phase. I used 2 X 4 just to keep the
visual clutter down. As shown, the clock is running at 1MHz.
The input signal is a single sharp spike, an approximation to the
Dirac Delta function, which produces a very wide spectrum.
The simulation output is selected as the voltage at the input winding
of the balun. As you might expect, it is also a single sharp spike.
The voltage at that node can be taken as a surrogate for the input
impedance of the detector.
The real treasure of the simulation is the FFT spectrum at that node,
shown. You can clearly see the intrinsic bandpass behavior of the
Tayloe circuit, as well as its receptivity at odd clock harmonics. As
the clock frequency changes, the response curve follows it. As the
Cacc parameter alters the values of the accumulating capacitors, the
bandwidth changes.
JZ KJ4A
toggle quoted message
Show quoted text
On Fri, Oct 20, 2023 at 3:17?AM Alan G4ZFQ <alan4alan@...> wrote:
On 19/10/2023 16:35, Rod Smith via groups.io wrote:
does this link work for you?
<>
Rod,
Yes, thanks.
An impressive project for it's time. Some parts can be improved and
simplified by modern developments.
The designers certainly believed in good filtering, no compromises.
73 Alan G4ZFQ
|
On 19/10/2023 16:35, Rod Smith via groups.io wrote: does this link work for you?
<>
Rod, Yes, thanks. An impressive project for it's time. Some parts can be improved and simplified by modern developments. The designers certainly believed in good filtering, no compromises. 73 Alan G4ZFQ
|
Hi Alan,
does this link work for you??
I haven¡¯t managed to get my head around how the Tayloe /QSD responds to frequencies below the signal frequency ¡ and found very little online about it. ?But I was aware of the 3 x signal frequency response - quite useful at higher frequencies perhaps.?
73
Rod
|
On 19/10/2023 12:19, Rod Smith via groups.io wrote: the main spurious responses
are:
1) the first sub-harmonic of the tuned frequency (e.g. unwanted reception of a strong signal
at 3.5MHz signal when tuned to 7.0MHz). Rejection of the first sub-harmonic is about
55 ¨C 60dB. Rod, After over 10 years of using Softrocks I'm surprised to read that. I'm not saying it is not a spurii although I've never seen it mentioned. What is always mentioned is the Tayloe's response to signals at odd multiples of the local oscillator. I reckon the biggest unwanted response is that of a signal at 3 times the tuned frequency. Without filtering it is 8-9dB lower than the wanted frequency. In fact may be used in order to keep the local oscillator frequency lower than the FST3253's maximum. Is the link live? I could not read it for myself, no response. Is there easy access somewhere? 73 Alan G4ZFQ
|
On 19/10/2023 07:52, Ted 2E0THH wrote: way you have titled this topic Stefan, I'm questioning the "Zero" because you do not display it. The waterfall and spectrum should show some noise to indicate that there are no signals. Completely black means nothing shows above a certain level. It is very rare not to see some weak traces on a waterfall adjusted for best sensitivity. As Hans and others say, testing a receiver is not simple. One thing that a properly tuned L401 does is help suppress interference from signals 3 times the tuned frequency. The mixer will be about 8dB down on a 3x signal. The LPF will reduce a strong signal quite a lot but testing at that frequency will demonstrate that a tuned L401 circuit helps further.. Actually Zero does not exist, in the beginning there was nothing, then along came a singularity that started it all:-) 73 Alan G4ZFQ
|
Ryuji Suzuki AB1WX
Peter Ayearst