I've tried to do a simple and low-cost sound-card PN system too back in the days. It's fun, when you see the broadband thermal resistor noise in your FFT plots :) But when doing experiments with the crystal oscillators, I've found out that one of the *MOST* important things to worry about is injection locking, especially if you do OCXOs and very low PN oscillators. Before feeding the RF and LO ports of the mixer, you should have isolation amps on both ports with about or more than 100dB S12 isolation, not to get into trouble. Without them, you'll be measuring unreal PN noise... Cheers, Leon

Here a small schematic of the simple Phase Noise Analyzer Hope this answers some of the questions. The +5 V reference from the OCXO is buffered by U1. The 3 ports of the used ADE-1 mixer are galvanically  isolated greatly helping to reduce ground loops. The output of the mixer is low passed using C2,L1 and C3 and used as input to the tuning of the OCXO R4/R4 are actually two potmeters linked with summing resistors for coarse/fine frequency adjustment. Inside the OCXO is a R/C low pass filter with a corner frequency of about 0.5 Hz. The potmeter setting do influence the loop gain but in practice this is not a problem. The mixer output is also amplified by U2 using automatic bias done with a large C1 and send into the audio input of a PC running the FFT program. The 10MHz output from the DUT goes into RF_GND and RF_IN For simplicity the supply decoupling capacitors and the output DC blocking capacitor are not drawn. The opamps use single +12 V, just like the OCXO. The OCXO must have better (or just equal) phase noise performance compared to the oscillator being measured (the DUT) The +12 V supply comes from a bench supply with floating ground By shorting C3 one can check for unwanted behavior like injection locking and measure the  internally generated noise. R5 is added to measure noise levels when no DUT is connected. The performance is surprisingly good, although one has to use a frequency counter to bring the DUT and OCXO close enough for lock. To check for 90 degrees lock the R3/R4 potmeters are tuned to maximum noise level while still having lock. Listening to the audio out is like listening to a DSB receiver. One can hear any disturbance or stray 10MHz. like the 10MHz house clock distribution cable being too close (not connected) to the Phase Noise Analyzer. Shielding is important to keep the noise down. By ensuring the FFT has a bandwidth of 1Hz a calibrated noise source can be use to establish a power level reference, much needed because of the undefined gains in the PC audio path. I know this design is far away from what many people on this list are used to, but it was good enough for me to quickly check the performance of some oscillators. Erik.

Hi If you need that sort of isolation, it certainly can be done. NIST has papers on very simple / DIY compatible cascode amps that will do the trick. ( chain of common base stages driven by a common emitter). Some folks on the list have gone a lot further in terms of complexity than NIST did. Device wise, the cascode amps seem to work pretty well with some very humble transistors ( 2N3904 etc ). There likely are fancier parts out there, but some of the really old stuff appears to be “good enough”. Why cascode ( or just common base) ? You can manage the gain pretty well. The amps likely are going to be quite low gain along with low noise. Thats not a great combination. There are a lot of OCXO’s out there that don’t need anything as crazy as the multi stage NIST amps. If you are looking at a system (as opposed to an oscillator) isolation already has been addressed in the instrument. Part of the “look at the beat note” process is to observe that it looks like a proper clipped triangle wave. The slopes on the rising and falling edges should be similar and of a reasonable slope. If you have injection locking ( and with a typical low frequency note) the rising and falling edges will begin to distort. The “lock” process will start to occur as the note crosses zero. This is yet another good reason to use a low frequency note. It also is good to look at the slope of *both* zero crossings. Bob > On Jul 5, 2022, at 12:16 AM, Leon Pavlovic via time-nuts <time-nuts@lists.febo.com> wrote: > > I've tried to do a simple and low-cost sound-card PN system too back in the > days. It's fun, when you see the broadband thermal resistor noise in your > FFT plots :) But when doing experiments with the crystal oscillators, I've > found out that one of the *MOST* important things to worry about is > injection locking, especially if you do OCXOs and very low PN oscillators. > > Before feeding the RF and LO ports of the mixer, you should have isolation > amps on both ports with about or more than 100dB S12 isolation, not to get > into trouble. Without them, you'll be measuring unreal PN noise... > > Cheers, > Leon > _______________________________________________ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe send an email to time-nuts-leave@lists.febo.com

Hi One “cute trick” that can be done on the lowpass filter: Ideally you would like to terminate the mixer properly at the DUT and DUT x 2 frequencies. Open circuit (or short circuit) is ok at audio. The quick and easy way to do this is to put an appropriate resistor ( 50 ohm, 500 ohm, whatever ) in series with C2 on your schematic. It is common to put an RF choke across it to get it “out of the way” as you head to audio. Spice is your friend in this case. It’s way easier to pump it into LTSpice or something similar and tinker then trying to come up with some sort of math solution. Stuffing in a resistor at the output of the lowpass is also a common thing. Again the value is a “that depends” sort of thing. 500 ohms or 5K are the likely candidates. The main value is to take the port to zero when the mixer is not being driven. No, none of this is a big deal. Since the ref out of the OCXO is likely a pretty noisy item, I would not get it anywhere near the low noise audio part of the circuit. The mixer output is typically grounded and the ref out is ignored. The EFC is driven off of the lock section of the circuit to keep things in quadrature. If things are “locking up” without a lock circuit then indeed the earlier post about injection locking applies. Bob > On Jul 5, 2022, at 3:15 AM, Erik Kaashoek via time-nuts <time-nuts@lists.febo.com> wrote: > > Here a small schematic of the simple Phase Noise Analyzer > Hope this answers some of the questions. > The +5 V reference from the OCXO is buffered by U1. > The 3 ports of the used ADE-1 mixer are galvanically isolated greatly helping to reduce ground loops. > The output of the mixer is low passed using C2,L1 and C3 and used as input to the tuning of the OCXO > R4/R4 are actually two potmeters linked with summing resistors for coarse/fine frequency adjustment. > Inside the OCXO is a R/C low pass filter with a corner frequency of about 0.5 Hz. > The potmeter setting do influence the loop gain but in practice this is not a problem. > The mixer output is also amplified by U2 using automatic bias done with a large C1 and send into the audio input of a PC running the FFT program. > The 10MHz output from the DUT goes into RF_GND and RF_IN > For simplicity the supply decoupling capacitors and the output DC blocking capacitor are not drawn. > The opamps use single +12 V, just like the OCXO. > The OCXO must have better (or just equal) phase noise performance compared to the oscillator being measured (the DUT) > The +12 V supply comes from a bench supply with floating ground > > By shorting C3 one can check for unwanted behavior like injection locking and measure the internally generated noise. > R5 is added to measure noise levels when no DUT is connected. > > The performance is surprisingly good, although one has to use a frequency counter to bring the DUT and OCXO close enough for lock. > To check for 90 degrees lock the R3/R4 potmeters are tuned to maximum noise level while still having lock. > Listening to the audio out is like listening to a DSB receiver. One can hear any disturbance or stray 10MHz. like the 10MHz house clock distribution cable being too close (not connected) to the Phase Noise Analyzer. > Shielding is important to keep the noise down. > By ensuring the FFT has a bandwidth of 1Hz a calibrated noise source can be use to establish a power level reference, much needed because of the undefined gains in the PC audio path. > > I know this design is far away from what many people on this list are used to, but it was good enough for me to quickly check the performance of some oscillators. > Erik.<PNA.pdf>_______________________________________________ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe send an email to time-nuts-leave@lists.febo.com

Bob, See below Thanks. Erik. On 5-7-2022 19:37, Bob kb8tq wrote: > Hi > > One “cute trick” that can be done on the lowpass filter: > > Ideally you would like to terminate the mixer properly at the DUT and DUT x 2 > frequencies. Open circuit (or short circuit) is ok at audio. The quick and easy way > to do this is to put an appropriate resistor ( 50 ohm, 500 ohm, whatever ) in series > with C2 on your schematic. It is common to put an RF choke across it to get it > “out of the way” as you head to audio. That is exactly what I did, forgot to show in schematic. > > Spice is your friend in this case. It’s way easier to pump it into LTSpice or something > similar and tinker then trying to come up with some sort of math solution. > > Stuffing in a resistor at the output of the lowpass is also a common thing. Again the > value is a “that depends” sort of thing. 500 ohms or 5K are the likely candidates. The > main value is to take the port to zero when the mixer is not being driven. I used 5k > > No, none of this is a big deal. > > Since the ref out of the OCXO is likely a pretty noisy item, I would not get it anywhere > near the low noise audio part of the circuit. The mixer output is typically grounded and > the ref out is ignored. There is an RC filter between the REF out and the buffer opamp, but I can move the mixer to ground and feed the bottom of the tune potentiometer (R3/R4), will test, but U2 still need 5V as reference for operation. An easier method may be to connect the auto bias capacitor C1 not to ground but to the buffered Vref, this puts the opamp into balanced input mode thus eliminating any remaining noise from Vref. > The EFC is driven off of the lock section of the circuit to keep > things in quadrature. > > If things are “locking up” without a lock circuit then indeed the earlier post about injection > locking applies. Will test for locking with mixer output shorted. > > Bob > >> On Jul 5, 2022, at 3:15 AM, Erik Kaashoek via time-nuts <time-nuts@lists.febo.com> wrote: >> >> Here a small schematic of the simple Phase Noise Analyzer >> Hope this answers some of the questions. >> The +5 V reference from the OCXO is buffered by U1. >> The 3 ports of the used ADE-1 mixer are galvanically isolated greatly helping to reduce ground loops. >> The output of the mixer is low passed using C2,L1 and C3 and used as input to the tuning of the OCXO >> R4/R4 are actually two potmeters linked with summing resistors for coarse/fine frequency adjustment. >> Inside the OCXO is a R/C low pass filter with a corner frequency of about 0.5 Hz. >> The potmeter setting do influence the loop gain but in practice this is not a problem. >> The mixer output is also amplified by U2 using automatic bias done with a large C1 and send into the audio input of a PC running the FFT program. >> The 10MHz output from the DUT goes into RF_GND and RF_IN >> For simplicity the supply decoupling capacitors and the output DC blocking capacitor are not drawn. >> The opamps use single +12 V, just like the OCXO. >> The OCXO must have better (or just equal) phase noise performance compared to the oscillator being measured (the DUT) >> The +12 V supply comes from a bench supply with floating ground >> >> By shorting C3 one can check for unwanted behavior like injection locking and measure the internally generated noise. >> R5 is added to measure noise levels when no DUT is connected. >> >> The performance is surprisingly good, although one has to use a frequency counter to bring the DUT and OCXO close enough for lock. >> To check for 90 degrees lock the R3/R4 potmeters are tuned to maximum noise level while still having lock. >> Listening to the audio out is like listening to a DSB receiver. One can hear any disturbance or stray 10MHz. like the 10MHz house clock distribution cable being too close (not connected) to the Phase Noise Analyzer. >> Shielding is important to keep the noise down. >> By ensuring the FFT has a bandwidth of 1Hz a calibrated noise source can be use to establish a power level reference, much needed because of the undefined gains in the PC audio path. >> >> I know this design is far away from what many people on this list are used to, but it was good enough for me to quickly check the performance of some oscillators. >> Erik.<PNA.pdf>_______________________________________________ >> time-nuts mailing list -- time-nuts@lists.febo.com >> To unsubscribe send an email to time-nuts-leave@lists.febo.com

Hi One of the “gotcha’s” with this is the need for dual supplies. That’s what lets you get to an 30 something volt peak to peak output on the initial amp. It also keeps all the signals ground referenced through the system. The downside ( as noted in posts from a number of years ago) is that sequencing of the supplies can matter. Having just one supply running and the other missing in action can put a bit of current where you might not want it to go. The common example is the mixer output port. Depending on how things are done, there are other possible paths. There are a lot of ways to deal with this. One quick check of “how am I doing?”: Look at the fundamental amplitude of the undistorted beat note. Then look at the 1 Hz normalized noise with both ports driven from the same device. Your noise should be more than 145 db down. The same should also be true of the preamp output with the input terminated in your magic 50 or 500 ohm resistance. Since the preamp likely does not pass the beat note without saturating, there is a bit of math here or there to get this all done. Is this gizmo going to have a flat floor right into 1 Hz? Nope. All these magic floor numbers are talking about >= 1 KHz off carrier. 1/F noise is a very real thing in your mixer, and in the op amp used for the preamp. Fortunately it also impacts signal sources. You are not likely to need -155 dbc / Hz data at 1Hz off carrier. Bob > On Jul 5, 2022, at 10:51 AM, Erik Kaashoek <erik@kaashoek.com> wrote: > > Bob, See below > Thanks. > Erik. > > On 5-7-2022 19:37, Bob kb8tq wrote: >> Hi >> >> One “cute trick” that can be done on the lowpass filter: >> >> Ideally you would like to terminate the mixer properly at the DUT and DUT x 2 >> frequencies. Open circuit (or short circuit) is ok at audio. The quick and easy way >> to do this is to put an appropriate resistor ( 50 ohm, 500 ohm, whatever ) in series >> with C2 on your schematic. It is common to put an RF choke across it to get it >> “out of the way” as you head to audio. > That is exactly what I did, forgot to show in schematic. >> >> Spice is your friend in this case. It’s way easier to pump it into LTSpice or something >> similar and tinker then trying to come up with some sort of math solution. >> >> Stuffing in a resistor at the output of the lowpass is also a common thing. Again the >> value is a “that depends” sort of thing. 500 ohms or 5K are the likely candidates. The >> main value is to take the port to zero when the mixer is not being driven. > I used 5k >> >> No, none of this is a big deal. >> >> Since the ref out of the OCXO is likely a pretty noisy item, I would not get it anywhere >> near the low noise audio part of the circuit. The mixer output is typically grounded and >> the ref out is ignored. > There is an RC filter between the REF out and the buffer opamp, but I can move the mixer to ground and feed the bottom of the tune potentiometer (R3/R4), will test, but U2 still need 5V as reference for operation. > An easier method may be to connect the auto bias capacitor C1 not to ground but to the buffered Vref, this puts the opamp into balanced input mode thus eliminating any remaining noise from Vref. >> The EFC is driven off of the lock section of the circuit to keep >> things in quadrature. >> >> If things are “locking up” without a lock circuit then indeed the earlier post about injection >> locking applies. > Will test for locking with mixer output shorted. >> >> Bob >> >>> On Jul 5, 2022, at 3:15 AM, Erik Kaashoek via time-nuts <time-nuts@lists.febo.com> wrote: >>> >>> Here a small schematic of the simple Phase Noise Analyzer >>> Hope this answers some of the questions. >>> The +5 V reference from the OCXO is buffered by U1. >>> The 3 ports of the used ADE-1 mixer are galvanically isolated greatly helping to reduce ground loops. >>> The output of the mixer is low passed using C2,L1 and C3 and used as input to the tuning of the OCXO >>> R4/R4 are actually two potmeters linked with summing resistors for coarse/fine frequency adjustment. >>> Inside the OCXO is a R/C low pass filter with a corner frequency of about 0.5 Hz. >>> The potmeter setting do influence the loop gain but in practice this is not a problem. >>> The mixer output is also amplified by U2 using automatic bias done with a large C1 and send into the audio input of a PC running the FFT program. >>> The 10MHz output from the DUT goes into RF_GND and RF_IN >>> For simplicity the supply decoupling capacitors and the output DC blocking capacitor are not drawn. >>> The opamps use single +12 V, just like the OCXO. >>> The OCXO must have better (or just equal) phase noise performance compared to the oscillator being measured (the DUT) >>> The +12 V supply comes from a bench supply with floating ground >>> >>> By shorting C3 one can check for unwanted behavior like injection locking and measure the internally generated noise. >>> R5 is added to measure noise levels when no DUT is connected. >>> >>> The performance is surprisingly good, although one has to use a frequency counter to bring the DUT and OCXO close enough for lock. >>> To check for 90 degrees lock the R3/R4 potmeters are tuned to maximum noise level while still having lock. >>> Listening to the audio out is like listening to a DSB receiver. One can hear any disturbance or stray 10MHz. like the 10MHz house clock distribution cable being too close (not connected) to the Phase Noise Analyzer. >>> Shielding is important to keep the noise down. >>> By ensuring the FFT has a bandwidth of 1Hz a calibrated noise source can be use to establish a power level reference, much needed because of the undefined gains in the PC audio path. >>> >>> I know this design is far away from what many people on this list are used to, but it was good enough for me to quickly check the performance of some oscillators. >>> Erik.<PNA.pdf>_______________________________________________ >>> time-nuts mailing list -- time-nuts@lists.febo.com >>> To unsubscribe send an email to time-nuts-leave@lists.febo.com >

Why would using a calibrated noise source into a 1Hz per bin FFT not be a good level check? Just subtract 3 dB due to both widebands folding and you have a reference. A noise source should also be great for checking flatness I think. A noise source at -90dBm/Hz around 10 MHz is cheap on eBay. Calibration can be done with any SA having noise markers. On Tue, Jul 5, 2022, 21:34 Bob kb8tq <kb8tq@n1k.org> wrote: > Hi > > One of the “gotcha’s” with this is the need for dual supplies. That’s what > lets you > get to an 30 something volt peak to peak output on the initial amp. It > also keeps > all the signals ground referenced through the system. > > The downside ( as noted in posts from a number of years ago) is that > sequencing > of the supplies can matter. Having just one supply running and the other > missing in > action can put a bit of current where you might not want it to go. The > common > example is the mixer output port. Depending on how things are done, there > are > other possible paths. There are a lot of ways to deal with this. > > One quick check of “how am I doing?”: Look at the fundamental amplitude of > the > undistorted beat note. Then look at the 1 Hz normalized noise with both > ports driven > from the same device. Your noise should be more than 145 db down. The same > should also be true of the preamp output with the input terminated in your > magic > 50 or 500 ohm resistance. Since the preamp likely does not pass the beat > note > without saturating, there is a bit of math here or there to get this all > done. > > Is this gizmo going to have a flat floor right into 1 Hz? Nope. All these > magic floor > numbers are talking about >= 1 KHz off carrier. 1/F noise is a very real > thing in > your mixer, and in the op amp used for the preamp. Fortunately it also > impacts > signal sources. You are not likely to need -155 dbc / Hz data at 1Hz off > carrier. > > Bob > > > On Jul 5, 2022, at 10:51 AM, Erik Kaashoek <erik@kaashoek.com> wrote: > > > > Bob, See below > > Thanks. > > Erik. > > > > On 5-7-2022 19:37, Bob kb8tq wrote: > >> Hi > >> > >> One “cute trick” that can be done on the lowpass filter: > >> > >> Ideally you would like to terminate the mixer properly at the DUT and > DUT x 2 > >> frequencies. Open circuit (or short circuit) is ok at audio. The quick > and easy way > >> to do this is to put an appropriate resistor ( 50 ohm, 500 ohm, > whatever ) in series > >> with C2 on your schematic. It is common to put an RF choke across it to > get it > >> “out of the way” as you head to audio. > > That is exactly what I did, forgot to show in schematic. > >> > >> Spice is your friend in this case. It’s way easier to pump it into > LTSpice or something > >> similar and tinker then trying to come up with some sort of math > solution. > >> > >> Stuffing in a resistor at the output of the lowpass is also a common > thing. Again the > >> value is a “that depends” sort of thing. 500 ohms or 5K are the likely > candidates. The > >> main value is to take the port to zero when the mixer is not being > driven. > > I used 5k > >> > >> No, none of this is a big deal. > >> > >> Since the ref out of the OCXO is likely a pretty noisy item, I would > not get it anywhere > >> near the low noise audio part of the circuit. The mixer output is > typically grounded and > >> the ref out is ignored. > > There is an RC filter between the REF out and the buffer opamp, but I > can move the mixer to ground and feed the bottom of the tune potentiometer > (R3/R4), will test, but U2 still need 5V as reference for operation. > > An easier method may be to connect the auto bias capacitor C1 not to > ground but to the buffered Vref, this puts the opamp into balanced input > mode thus eliminating any remaining noise from Vref. > >> The EFC is driven off of the lock section of the circuit to keep > >> things in quadrature. > >> > >> If things are “locking up” without a lock circuit then indeed the > earlier post about injection > >> locking applies. > > Will test for locking with mixer output shorted. > >> > >> Bob > >> > >>> On Jul 5, 2022, at 3:15 AM, Erik Kaashoek via time-nuts < > time-nuts@lists.febo.com> wrote: > >>> > >>> Here a small schematic of the simple Phase Noise Analyzer > >>> Hope this answers some of the questions. > >>> The +5 V reference from the OCXO is buffered by U1. > >>> The 3 ports of the used ADE-1 mixer are galvanically isolated greatly > helping to reduce ground loops. > >>> The output of the mixer is low passed using C2,L1 and C3 and used as > input to the tuning of the OCXO > >>> R4/R4 are actually two potmeters linked with summing resistors for > coarse/fine frequency adjustment. > >>> Inside the OCXO is a R/C low pass filter with a corner frequency of > about 0.5 Hz. > >>> The potmeter setting do influence the loop gain but in practice this > is not a problem. > >>> The mixer output is also amplified by U2 using automatic bias done > with a large C1 and send into the audio input of a PC running the FFT > program. > >>> The 10MHz output from the DUT goes into RF_GND and RF_IN > >>> For simplicity the supply decoupling capacitors and the output DC > blocking capacitor are not drawn. > >>> The opamps use single +12 V, just like the OCXO. > >>> The OCXO must have better (or just equal) phase noise performance > compared to the oscillator being measured (the DUT) > >>> The +12 V supply comes from a bench supply with floating ground > >>> > >>> By shorting C3 one can check for unwanted behavior like injection > locking and measure the internally generated noise. > >>> R5 is added to measure noise levels when no DUT is connected. > >>> > >>> The performance is surprisingly good, although one has to use a > frequency counter to bring the DUT and OCXO close enough for lock. > >>> To check for 90 degrees lock the R3/R4 potmeters are tuned to maximum > noise level while still having lock. > >>> Listening to the audio out is like listening to a DSB receiver. One > can hear any disturbance or stray 10MHz. like the 10MHz house clock > distribution cable being too close (not connected) to the Phase Noise > Analyzer. > >>> Shielding is important to keep the noise down. > >>> By ensuring the FFT has a bandwidth of 1Hz a calibrated noise source > can be use to establish a power level reference, much needed because of the > undefined gains in the PC audio path. > >>> > >>> I know this design is far away from what many people on this list are > used to, but it was good enough for me to quickly check the performance of > some oscillators. > >>> Erik.<PNA.pdf>_______________________________________________ > >>> time-nuts mailing list -- time-nuts@lists.febo.com > >>> To unsubscribe send an email to time-nuts-leave@lists.febo.com > > > >

Hi The mixer sensitivity is based on both ports being saturated. You are trying to check the circuit end to end in it’s “as used” configuration. There is no real need for an added calibration source, provided you pay attention to the basics. Bob > On Jul 5, 2022, at 12:04 PM, Erik Kaashoek <erik@kaashoek.com> wrote: > > Why would using a calibrated noise source into a 1Hz per bin FFT not be a good level check? Just subtract 3 dB due to both widebands folding and you have a reference. > A noise source should also be great for checking flatness I think. > A noise source at -90dBm/Hz around 10 MHz is cheap on eBay. Calibration can be done with any SA having noise markers. > > On Tue, Jul 5, 2022, 21:34 Bob kb8tq <kb8tq@n1k.org <mailto:kb8tq@n1k.org>> wrote: > Hi > > One of the “gotcha’s” with this is the need for dual supplies. That’s what lets you > get to an 30 something volt peak to peak output on the initial amp. It also keeps > all the signals ground referenced through the system. > > The downside ( as noted in posts from a number of years ago) is that sequencing > of the supplies can matter. Having just one supply running and the other missing in > action can put a bit of current where you might not want it to go. The common > example is the mixer output port. Depending on how things are done, there are > other possible paths. There are a lot of ways to deal with this. > > One quick check of “how am I doing?”: Look at the fundamental amplitude of the > undistorted beat note. Then look at the 1 Hz normalized noise with both ports driven > from the same device. Your noise should be more than 145 db down. The same > should also be true of the preamp output with the input terminated in your magic > 50 or 500 ohm resistance. Since the preamp likely does not pass the beat note > without saturating, there is a bit of math here or there to get this all done. > > Is this gizmo going to have a flat floor right into 1 Hz? Nope. All these magic floor > numbers are talking about >= 1 KHz off carrier. 1/F noise is a very real thing in > your mixer, and in the op amp used for the preamp. Fortunately it also impacts > signal sources. You are not likely to need -155 dbc / Hz data at 1Hz off carrier. > > Bob > > > On Jul 5, 2022, at 10:51 AM, Erik Kaashoek <erik@kaashoek.com <mailto:erik@kaashoek.com>> wrote: > > > > Bob, See below > > Thanks. > > Erik. > > > > On 5-7-2022 19:37, Bob kb8tq wrote: > >> Hi > >> > >> One “cute trick” that can be done on the lowpass filter: > >> > >> Ideally you would like to terminate the mixer properly at the DUT and DUT x 2 > >> frequencies. Open circuit (or short circuit) is ok at audio. The quick and easy way > >> to do this is to put an appropriate resistor ( 50 ohm, 500 ohm, whatever ) in series > >> with C2 on your schematic. It is common to put an RF choke across it to get it > >> “out of the way” as you head to audio. > > That is exactly what I did, forgot to show in schematic. > >> > >> Spice is your friend in this case. It’s way easier to pump it into LTSpice or something > >> similar and tinker then trying to come up with some sort of math solution. > >> > >> Stuffing in a resistor at the output of the lowpass is also a common thing. Again the > >> value is a “that depends” sort of thing. 500 ohms or 5K are the likely candidates. The > >> main value is to take the port to zero when the mixer is not being driven. > > I used 5k > >> > >> No, none of this is a big deal. > >> > >> Since the ref out of the OCXO is likely a pretty noisy item, I would not get it anywhere > >> near the low noise audio part of the circuit. The mixer output is typically grounded and > >> the ref out is ignored. > > There is an RC filter between the REF out and the buffer opamp, but I can move the mixer to ground and feed the bottom of the tune potentiometer (R3/R4), will test, but U2 still need 5V as reference for operation. > > An easier method may be to connect the auto bias capacitor C1 not to ground but to the buffered Vref, this puts the opamp into balanced input mode thus eliminating any remaining noise from Vref. > >> The EFC is driven off of the lock section of the circuit to keep > >> things in quadrature. > >> > >> If things are “locking up” without a lock circuit then indeed the earlier post about injection > >> locking applies. > > Will test for locking with mixer output shorted. > >> > >> Bob > >> > >>> On Jul 5, 2022, at 3:15 AM, Erik Kaashoek via time-nuts <time-nuts@lists.febo.com <mailto:time-nuts@lists.febo.com>> wrote: > >>> > >>> Here a small schematic of the simple Phase Noise Analyzer > >>> Hope this answers some of the questions. > >>> The +5 V reference from the OCXO is buffered by U1. > >>> The 3 ports of the used ADE-1 mixer are galvanically isolated greatly helping to reduce ground loops. > >>> The output of the mixer is low passed using C2,L1 and C3 and used as input to the tuning of the OCXO > >>> R4/R4 are actually two potmeters linked with summing resistors for coarse/fine frequency adjustment. > >>> Inside the OCXO is a R/C low pass filter with a corner frequency of about 0.5 Hz. > >>> The potmeter setting do influence the loop gain but in practice this is not a problem. > >>> The mixer output is also amplified by U2 using automatic bias done with a large C1 and send into the audio input of a PC running the FFT program. > >>> The 10MHz output from the DUT goes into RF_GND and RF_IN > >>> For simplicity the supply decoupling capacitors and the output DC blocking capacitor are not drawn. > >>> The opamps use single +12 V, just like the OCXO. > >>> The OCXO must have better (or just equal) phase noise performance compared to the oscillator being measured (the DUT) > >>> The +12 V supply comes from a bench supply with floating ground > >>> > >>> By shorting C3 one can check for unwanted behavior like injection locking and measure the internally generated noise. > >>> R5 is added to measure noise levels when no DUT is connected. > >>> > >>> The performance is surprisingly good, although one has to use a frequency counter to bring the DUT and OCXO close enough for lock. > >>> To check for 90 degrees lock the R3/R4 potmeters are tuned to maximum noise level while still having lock. > >>> Listening to the audio out is like listening to a DSB receiver. One can hear any disturbance or stray 10MHz. like the 10MHz house clock distribution cable being too close (not connected) to the Phase Noise Analyzer. > >>> Shielding is important to keep the noise down. > >>> By ensuring the FFT has a bandwidth of 1Hz a calibrated noise source can be use to establish a power level reference, much needed because of the undefined gains in the PC audio path. > >>> > >>> I know this design is far away from what many people on this list are used to, but it was good enough for me to quickly check the performance of some oscillators. > >>> Erik.<PNA.pdf>_______________________________________________ > >>> time-nuts mailing list -- time-nuts@lists.febo.com <mailto:time-nuts@lists.febo.com> > >>> To unsubscribe send an email to time-nuts-leave@lists.febo.com <mailto:time-nuts-leave@lists.febo.com> > > >

Am 2022-07-05 18:04, schrieb Bob kb8tq via time-nuts: > If you need that sort of isolation, it certainly can be done. > NIST has papers on very simple / DIY compatible cascode > amps that will do the trick. ( chain of common base stages > driven by a common emitter). Some folks on the list have > gone a lot further in terms of complexity than NIST did. > > Device wise, the cascode amps seem to work pretty well > with some very humble transistors ( 2N3904 etc ). There > likely are fancier parts out there, but some of the really > old stuff appears to be “good enough”. I have made a new isolation amplifier but I'm absolutely not happy with the available transistors. Anything in sot-89 is either to slow ( Zetex/Diodes Inc, the 2N3904-alikes) or is much too hot. I want at least 200 MHz to have no phase shift at 100. BFQ19s gave me 1 GHz of BW. The version in the plot is already heavily sandbaged but still has quite an S21 overshoot on the high frequency end. The input-voltage to cascode current converter is especially problematic in that the smallest capacitive load on the emitter tends to make it more unstable. That spoils S11, of course. I even took the feedback from a tap of the emitter resistor. Backward isolation is 120 dB over most of the useful range but changes depending on the damping methods. Any ideas of more friendly transistors? BFQ31 were quite well-behaved but are extinct now. I still have a reel, but stuff from the secret drawer is unfair. And it's PNP. cheers, Gerhard

Some more testing done. The mixer starts to saturate at -3dBm and is fully saturated a -1dBm so for every input larger than -1dBm the phase demodulation output is no longer depending on the amplitude of the input signal, only on the phase modulation. It took a long time to hunt down the source of all spurs when no DUT was connected, every single 10MHz source within 2 meters had to be switched off or moved much farther away. Maybe time to start improving shielding, including cables. In the end the noise floor is -135 dBm/Hz between 10 and 100Hz offset, dropping down to -150dBm/Hz at 1kHz offset and above (see attached PN_baseline_2.JPG, subtract 40dBm from indicated levels) Above 1kHz there are still a large amount of spurs but these can easily be ignored. Something weird is happening at 20Hz and 25Hz offset. This could be due to the temperature control in the used DOXCO Measuring a not so good OXCO gave good results down to about 5 Hz offset (see PN_OCXO_2.JPG ) Locking now happens when the DUT and DOCXO are within 0.5Hz, it does take up to 10 seconds so both sources must be stable and not drifting due to warming up. The attached schematic has been updated. (PNA.pdf) showing the balanced input of U2 greatly helping to remove any noise from the REF voltage. Also shown is the 50 ohm resistor acting as sink for all the high frequency signals through the mixer. The phase noise analyzer is now in a small box and ready for the real world. Erik.