Hi Time-nuts people, I'm new to this forum and really enjoyed reading all the design strategies on oscillators. I'm currently designing a low noise oscillator and I wonder if you can help me with some questions. I've attached a picture of the oscillator core I'm using. I decided to go for a Pierce design some time ago after looking at a few topologies. I took a number of measures to give the design a low 1/f noise. 1. Low DC gain. There is a transformer in the drain line. 2. Ultra low noise supply (not on the schematic) using an ad586 buried zener reference with a -3dB lowpass at 0.1Hz. This 5V reference is then increased to 12V using a low noise discrete buffer. 3. The phase shift across T1 is only a few degrees (<5degrees), low phaseshift in the amplifier reduces 1/f noise. 4. I chose a fet in order to minimize the load at the 'output' of the crystal, the only load I have now is the bias resistor of 100k. This resistor does cause low frequency noise (at higher frequencies C1 shorts it). This may be the problem in this circuit, Cgd is modulated by this noise and a low frequency voltage is applied at the gate, which is however not amplified because of the currentsource in the source line which should cause nearly infinite feedback for low frequencies making the DC gain even lower. 5. The current source produces low frequency noise. I have to have a current source though (I could use a resistor but I calculated that produces more noise). I could increase the voltage across R5 and make R5 bigger to reduce the noise, however, this will reduce Vgd which means modulation of Cgd becomes worse. I think I'm ok here by dividing the voltage across the currentsource and the active oscillator element in half. Cgd of a J309 is around 2.5pF at Vds=10V (it is a little higher in my case since Vgd is lower) 6. L2 can be mounted to accommodate overtone crystals. ( I still have to calculate a value for L2 and C2 to get the correct impedance at resonance) 7. F1 and F2 are ferrite beads with a low impedance at DC (far less then an ohm) and rising impedance at higher frequencies to prevent oscillation of the RF transistors. 8. I've found a really good overtonecrystal from Citizen. CM309S. I measured the unloaded Q to be 313000, the loaded Q (using simple estimation) should be around 280000 in the circuit. (C0=2.5pF, R=20Ohm, C1=0.75fF, L=29.444mH, can be aquired at digikey). The spec for Rs was <130Ohms, I guess reality is much better :-) 9. The transformeroutputs are going to an isolationchain using 3 cascodes and then a discrete limiter not shown on the schematic. 10. This whole oscillator core will be placed in an RF shielding can. The isolationchain will have a can of it's own to so pulling should be nearly eliminated. Now my questions are: 1. I don't see a pierce design in any of the low noise oscillator circuits in the discussion threads about low noise oscillator design. Is there something fundamentally wrong about this topology? 2. I read a few times that ferrites in inductors (and I assume transformers as well) can be a real problem for 1/f noise. I'm using a transformer (TC1-1t from minicircuits) which is made with a ferrite bead. Do you know of any transformers or inductors that have low 1/f noise ferrite material. 3. Have I missed something fundamental in the design. The goal is to build a very good oscillator. I would like to achieve something like -110dBc/Hz at 10Hz distance at 33.8688MHz. Thanks in advance for having a look and best regards, Hans Rosenberg

My best crystal oscillator designs are modified Pierce. But I definitely recommend using a low noise BJT else you will have 1/f noise. Jeffrey Pawlan WA6KBL Pawlan Communications

Hans Hans Rosenberg wrote: > Hi Time-nuts people, > > I'm new to this forum and really enjoyed reading all the design strategies on oscillators. > > I'm currently designing a low noise oscillator and I wonder if you can help me with some questions. I've attached a picture of the oscillator core I'm using. I decided to go for a Pierce design some time ago after looking at a few topologies. I took a number of measures to give the design a low 1/f noise. > > > 1. Low DC gain. There is a transformer in the drain line. > 2. Ultra low noise supply (not on the schematic) using an ad586 buried zener reference with a -3dB lowpass at 0.1Hz. This 5V reference is then increased to 12V using a low noise discrete buffer. > 3. The phase shift across T1 is only a few degrees (<5degrees), low phaseshift in the amplifier reduces 1/f noise. > 4. I chose a fet in order to minimize the load at the 'output' of the crystal, the only load I have now is the bias resistor of 100k. This resistor does cause low frequency noise (at higher frequencies C1 shorts it). This may be the problem in this circuit, Cgd is modulated by this noise and a low frequency voltage is applied at the gate, which is however not amplified because of the currentsource in the source line which should cause nearly infinite feedback for low frequencies making the DC gain even lower. > 5. The current source produces low frequency noise. I have to have a current source though (I could use a resistor but I calculated that produces more noise). I could increase the voltage across R5 and make R5 bigger to reduce the noise, however, this will reduce Vgd which means modulation of Cgd becomes worse. I think I'm ok here by dividing the voltage across the currentsource and the active oscillator element in half. Cgd of a J309 is around 2.5pF at Vds=10V (it is a little higher in my case since Vgd is lower) > 6. L2 can be mounted to accommodate overtone crystals. ( I still have to calculate a value for L2 and C2 to get the correct impedance at resonance) > 7. F1 and F2 are ferrite beads with a low impedance at DC (far less then an ohm) and rising impedance at higher frequencies to prevent oscillation of the RF transistors. > 8. I've found a really good overtonecrystal from Citizen. CM309S. I measured the unloaded Q to be 313000, the loaded Q (using simple estimation) should be around 280000 in the circuit. (C0=2.5pF, R=20Ohm, C1=0.75fF, L=29.444mH, can be aquired at digikey). The spec for Rs was <130Ohms, I guess reality is much better :-) > 9. The transformeroutputs are going to an isolationchain using 3 cascodes and then a discrete limiter not shown on the schematic. > 10. This whole oscillator core will be placed in an RF shielding can. The isolationchain will have a can of it's own to so pulling should be nearly eliminated. > > Now my questions are: > > > 1. I don't see a pierce design in any of the low noise oscillator circuits in the discussion threads about low noise oscillator design. Is there something fundamentally wrong about this topology? > 2. I read a few times that ferrites in inductors (and I assume transformers as well) can be a real problem for 1/f noise. I'm using a transformer (TC1-1t from minicircuits) which is made with a ferrite bead. Do you know of any transformers or inductors that have low 1/f noise ferrite material. > 3. Have I missed something fundamental in the design. The goal is to build a very good oscillator. I would like to achieve something like -110dBc/Hz at 10Hz distance at 33.8688MHz. > > Thanks in advance for having a look and best regards, > > Hans Rosenberg > > 1) A LED plus a BJT and resistor makes a much lower noise current source than a zener based reference plus a BJT and resistor. Some shielding of the LED from ambient light may be required as LEDs (as all PN junctions) are photosensitive. A resistor in series with an RF choke can be used to replace the current source. 2) A BJT with a correctly proportioned coupling network will not significantly load the crystal. 3) There is little isolation between the oscillator and the load, a slight change in topology will improve this. 4) Modified Pierce overtone crystal oscillators using JFETs were popular in the 1960's and 1970's. 5) The amplitude limiting mechanism in the oscillator is important as it affects the phase noise. 6) A lower noise audio BJT will make a much lower (flicker) noise current source than a 9GHz transistor. 7) Its usually better to use the crystal to filter the output signal. 8) One of the Driscoll BJT oscillators or a variant thereof is a good stating point. AGC via a varactor based attenuator can be used. The output signal can be extracted from the second transistor collector using a transformer with little interaction with the collector tank tuning. Bruce

Bruce Griffiths wrote: > Hans > > Hans Rosenberg wrote: > >> Hi Time-nuts people, >> >> I'm new to this forum and really enjoyed reading all the design strategies on oscillators. >> >> I'm currently designing a low noise oscillator and I wonder if you can help me with some questions. I've attached a picture of the oscillator core I'm using. I decided to go for a Pierce design some time ago after looking at a few topologies. I took a number of measures to give the design a low 1/f noise. >> >> >> 1. Low DC gain. There is a transformer in the drain line. >> 2. Ultra low noise supply (not on the schematic) using an ad586 buried zener reference with a -3dB lowpass at 0.1Hz. This 5V reference is then increased to 12V using a low noise discrete buffer. >> 3. The phase shift across T1 is only a few degrees (<5degrees), low phaseshift in the amplifier reduces 1/f noise. >> 4. I chose a fet in order to minimize the load at the 'output' of the crystal, the only load I have now is the bias resistor of 100k. This resistor does cause low frequency noise (at higher frequencies C1 shorts it). This may be the problem in this circuit, Cgd is modulated by this noise and a low frequency voltage is applied at the gate, which is however not amplified because of the currentsource in the source line which should cause nearly infinite feedback for low frequencies making the DC gain even lower. >> 5. The current source produces low frequency noise. I have to have a current source though (I could use a resistor but I calculated that produces more noise). I could increase the voltage across R5 and make R5 bigger to reduce the noise, however, this will reduce Vgd which means modulation of Cgd becomes worse. I think I'm ok here by dividing the voltage across the currentsource and the active oscillator element in half. Cgd of a J309 is around 2.5pF at Vds=10V (it is a little higher in my case since Vgd is lower) >> 6. L2 can be mounted to accommodate overtone crystals. ( I still have to calculate a value for L2 and C2 to get the correct impedance at resonance) >> 7. F1 and F2 are ferrite beads with a low impedance at DC (far less then an ohm) and rising impedance at higher frequencies to prevent oscillation of the RF transistors. >> 8. I've found a really good overtonecrystal from Citizen. CM309S. I measured the unloaded Q to be 313000, the loaded Q (using simple estimation) should be around 280000 in the circuit. (C0=2.5pF, R=20Ohm, C1=0.75fF, L=29.444mH, can be aquired at digikey). The spec for Rs was <130Ohms, I guess reality is much better :-) >> 9. The transformeroutputs are going to an isolationchain using 3 cascodes and then a discrete limiter not shown on the schematic. >> 10. This whole oscillator core will be placed in an RF shielding can. The isolationchain will have a can of it's own to so pulling should be nearly eliminated. >> >> Now my questions are: >> >> >> 1. I don't see a pierce design in any of the low noise oscillator circuits in the discussion threads about low noise oscillator design. Is there something fundamentally wrong about this topology? >> 2. I read a few times that ferrites in inductors (and I assume transformers as well) can be a real problem for 1/f noise. I'm using a transformer (TC1-1t from minicircuits) which is made with a ferrite bead. Do you know of any transformers or inductors that have low 1/f noise ferrite material. >> 3. Have I missed something fundamental in the design. The goal is to build a very good oscillator. I would like to achieve something like -110dBc/Hz at 10Hz distance at 33.8688MHz. >> >> Thanks in advance for having a look and best regards, >> >> Hans Rosenberg >> >> >> > 1) A LED plus a BJT and resistor makes a much lower noise current source > than a zener based reference plus a BJT and resistor. > Some shielding of the LED from ambient light may be required as LEDs (as > all PN junctions) are photosensitive. > A resistor in series with an RF choke can be used to replace the current > source. > > 2) A BJT with a correctly proportioned coupling network will not > significantly load the crystal. > > 3) There is little isolation between the oscillator and the load, a > slight change in topology will improve this. > > 4) Modified Pierce overtone crystal oscillators using JFETs were popular > in the 1960's and 1970's. > > 5) The amplitude limiting mechanism in the oscillator is important as it > affects the phase noise. > > 6) A lower noise audio BJT will make a much lower (flicker) noise > current source than a 9GHz transistor. > > 7) Its usually better to use the crystal to filter the output signal. > > 8) One of the Driscoll BJT oscillators or a variant thereof is a good > stating point. > AGC via a varactor based attenuator can be used. > The output signal can be extracted from the second transistor collector > using a transformer with little interaction with the collector tank tuning. > > > Bruce > > Addendum 9) Having a dc voltage across the crystal isnt a good idea as it can lead to unwanted phase modulation. 10) The current source transistor collector base voltage seems a little low. Bruce

Bruce Griffiths wrote: > Bruce Griffiths wrote: > >> Hans >> >> Hans Rosenberg wrote: >> >> >>> Hi Time-nuts people, >>> >>> I'm new to this forum and really enjoyed reading all the design strategies on oscillators. >>> >>> I'm currently designing a low noise oscillator and I wonder if you can help me with some questions. I've attached a picture of the oscillator core I'm using. I decided to go for a Pierce design some time ago after looking at a few topologies. I took a number of measures to give the design a low 1/f noise. >>> >>> >>> 1. Low DC gain. There is a transformer in the drain line. >>> 2. Ultra low noise supply (not on the schematic) using an ad586 buried zener reference with a -3dB lowpass at 0.1Hz. This 5V reference is then increased to 12V using a low noise discrete buffer. >>> 3. The phase shift across T1 is only a few degrees (<5degrees), low phaseshift in the amplifier reduces 1/f noise. >>> 4. I chose a fet in order to minimize the load at the 'output' of the crystal, the only load I have now is the bias resistor of 100k. This resistor does cause low frequency noise (at higher frequencies C1 shorts it). This may be the problem in this circuit, Cgd is modulated by this noise and a low frequency voltage is applied at the gate, which is however not amplified because of the currentsource in the source line which should cause nearly infinite feedback for low frequencies making the DC gain even lower. >>> 5. The current source produces low frequency noise. I have to have a current source though (I could use a resistor but I calculated that produces more noise). I could increase the voltage across R5 and make R5 bigger to reduce the noise, however, this will reduce Vgd which means modulation of Cgd becomes worse. I think I'm ok here by dividing the voltage across the currentsource and the active oscillator element in half. Cgd of a J309 is around 2.5pF at Vds=10V (it is a little higher in my case since Vgd is lower) >>> 6. L2 can be mounted to accommodate overtone crystals. ( I still have to calculate a value for L2 and C2 to get the correct impedance at resonance) >>> 7. F1 and F2 are ferrite beads with a low impedance at DC (far less then an ohm) and rising impedance at higher frequencies to prevent oscillation of the RF transistors. >>> 8. I've found a really good overtonecrystal from Citizen. CM309S. I measured the unloaded Q to be 313000, the loaded Q (using simple estimation) should be around 280000 in the circuit. (C0=2.5pF, R=20Ohm, C1=0.75fF, L=29.444mH, can be aquired at digikey). The spec for Rs was <130Ohms, I guess reality is much better :-) >>> 9. The transformeroutputs are going to an isolationchain using 3 cascodes and then a discrete limiter not shown on the schematic. >>> 10. This whole oscillator core will be placed in an RF shielding can. The isolationchain will have a can of it's own to so pulling should be nearly eliminated. >>> >>> Now my questions are: >>> >>> >>> 1. I don't see a pierce design in any of the low noise oscillator circuits in the discussion threads about low noise oscillator design. Is there something fundamentally wrong about this topology? >>> 2. I read a few times that ferrites in inductors (and I assume transformers as well) can be a real problem for 1/f noise. I'm using a transformer (TC1-1t from minicircuits) which is made with a ferrite bead. Do you know of any transformers or inductors that have low 1/f noise ferrite material. >>> 3. Have I missed something fundamental in the design. The goal is to build a very good oscillator. I would like to achieve something like -110dBc/Hz at 10Hz distance at 33.8688MHz. >>> >>> Thanks in advance for having a look and best regards, >>> >>> Hans Rosenberg >>> >>> >>> >>> >> 1) A LED plus a BJT and resistor makes a much lower noise current source >> than a zener based reference plus a BJT and resistor. >> Some shielding of the LED from ambient light may be required as LEDs (as >> all PN junctions) are photosensitive. >> A resistor in series with an RF choke can be used to replace the current >> source. >> >> 2) A BJT with a correctly proportioned coupling network will not >> significantly load the crystal. >> >> 3) There is little isolation between the oscillator and the load, a >> slight change in topology will improve this. >> >> 4) Modified Pierce overtone crystal oscillators using JFETs were popular >> in the 1960's and 1970's. >> >> 5) The amplitude limiting mechanism in the oscillator is important as it >> affects the phase noise. >> >> 6) A lower noise audio BJT will make a much lower (flicker) noise >> current source than a 9GHz transistor. >> >> 7) Its usually better to use the crystal to filter the output signal. >> >> 8) One of the Driscoll BJT oscillators or a variant thereof is a good >> stating point. >> AGC via a varactor based attenuator can be used. >> The output signal can be extracted from the second transistor collector >> using a transformer with little interaction with the collector tank tuning. >> >> >> Bruce >> >> >> > Addendum > > 9) Having a dc voltage across the crystal isnt a good idea as it can > lead to unwanted phase modulation. > > > 10) The current source transistor collector base voltage seems a little low. > > Bruce > > Addendum II 11) A common base cascade will have much lower noise than a cascode chain, however the CB cascade needs a current input. This can be derived in your modified Pierce circuit by connecting C1 to the emitter of the input CB transistor rather than to ground. The transformer is then not required. A lower impedance input at the CB stage emitter can be achieved by replacing the Cb transistor by a CECB feedback pair. Whats the limiter circuit? Is it at the cascode buffer chain output? Bruce

Hi Bruce, Thanks a lot for all the input. Do you have any figures on the noise voltage on a led? A good buried zener reference does 100nV/rt Hz, what does a led do? You mention that an amplitude control mechanism influences phase noise. Are you suggesting that a good amplitude control is better than turning off the active device for some time each cycle? I started having a close look at the driscoll design. It looks really nice. It seems much more elegant than what I was trying to do. I do have one question about the AGC circuit. The tune-voltage must be kept in the middle of the supply, otherwise the amplitude control loop will modulate the overall capacitance of the capacitive attenuator. So this is critical to set correctly in practice. There probably is a good reason why the current in the transistor is not changed to achieve amplitude control. Do the capacitances in the transistors vary so much that they start introducing more phase errors than when tuning a capacitive attenuator correctly? Or is there another reason. My isolation chain is indeed build with cb stages and not cascodes (wrong terminology from my side). I generate a current using a degenerated differential pair (the solution in the driscoll oscillator looks much more elegant). My limiterstage is build with a differential pair with 2 resistors in the collectors. The differential limited voltage goes to a transformer. The secondary side of that transformer is biased to 1.4V and drives a 74ABT inverter. I choose these inverters for their very low output impedance so I can drive a load with a high slewrate. However too much slewrate may cause a lot of interference and maybe reflections, so I'm thinking about putting a 33OHm resistor in series with the output to limit the bandwidth and reflections somewhat. Do you have any experience which family of ports (ac, hc, abt etc) delivers the best phase noise results (I should not have that much problems since I'm driving it with a pretty good squarewave but am curious what you think of it). I also would really like to get my hands on a 33.8688MHz SC-cut crystal, but I found they are hard to get. I guess this means starting up a production run at a manufacturer which is quite expensive. And I guess I'd have to include a temperature control circuit since the temperature characteristics of SC-cut crystals are not that good. Nice to hear that you're into optics as well! I build some basic telescopes, although I never made my own lenses or mirrors. But I'm very fascinated by the subject. I used to work for ASTRON, which is a company that builds radiotelescopes. I also used to give stargazing evenings for tourists at a local observatory which was a lot of fun. We had a 60cm mirrortelescope there. Unfortunately, this is a bit of overkill in holland with the bad seeing we usually have here. Thanks again for the nice input. Best regards, Hans Rosenberg ________________________________________ From: time-nuts-bounces@febo.com [time-nuts-bounces@febo.com] On Behalf Of Bruce Griffiths [bruce.griffiths@xtra.co.nz] Sent: Wednesday, January 14, 2009 10:41 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Ultra low noise Pierce oscillator??? Bruce Griffiths wrote: > Bruce Griffiths wrote: > >> Hans >> >> Hans Rosenberg wrote: >> >> >>> Hi Time-nuts people, >>> >>> I'm new to this forum and really enjoyed reading all the design strategies on oscillators. >>> >>> I'm currently designing a low noise oscillator and I wonder if you can help me with some questions. I've attached a picture of the oscillator core I'm using. I decided to go for a Pierce design some time ago after looking at a few topologies. I took a number of measures to give the design a low 1/f noise. >>> >>> >>> 1. Low DC gain. There is a transformer in the drain line. >>> 2. Ultra low noise supply (not on the schematic) using an ad586 buried zener reference with a -3dB lowpass at 0.1Hz. This 5V reference is then increased to 12V using a low noise discrete buffer. >>> 3. The phase shift across T1 is only a few degrees (<5degrees), low phaseshift in the amplifier reduces 1/f noise. >>> 4. I chose a fet in order to minimize the load at the 'output' of the crystal, the only load I have now is the bias resistor of 100k. This resistor does cause low frequency noise (at higher frequencies C1 shorts it). This may be the problem in this circuit, Cgd is modulated by this noise and a low frequency voltage is applied at the gate, which is however not amplified because of the currentsource in the source line which should cause nearly infinite feedback for low frequencies making the DC gain even lower. >>> 5. The current source produces low frequency noise. I have to have a current source though (I could use a resistor but I calculated that produces more noise). I could increase the voltage across R5 and make R5 bigger to reduce the noise, however, this will reduce Vgd which means modulation of Cgd becomes worse. I think I'm ok here by dividing the voltage across the currentsource and the active oscillator element in half. Cgd of a J309 is around 2.5pF at Vds=10V (it is a little higher in my case since Vgd is lower) >>> 6. L2 can be mounted to accommodate overtone crystals. ( I still have to calculate a value for L2 and C2 to get the correct impedance at resonance) >>> 7. F1 and F2 are ferrite beads with a low impedance at DC (far less then an ohm) and rising impedance at higher frequencies to prevent oscillation of the RF transistors. >>> 8. I've found a really good overtonecrystal from Citizen. CM309S. I measured the unloaded Q to be 313000, the loaded Q (using simple estimation) should be around 280000 in the circuit. (C0=2.5pF, R=20Ohm, C1=0.75fF, L=29.444mH, can be aquired at digikey). The spec for Rs was <130Ohms, I guess reality is much better :-) >>> 9. The transformeroutputs are going to an isolationchain using 3 cascodes and then a discrete limiter not shown on the schematic. >>> 10. This whole oscillator core will be placed in an RF shielding can. The isolationchain will have a can of it's own to so pulling should be nearly eliminated. >>> >>> Now my questions are: >>> >>> >>> 1. I don't see a pierce design in any of the low noise oscillator circuits in the discussion threads about low noise oscillator design. Is there something fundamentally wrong about this topology? >>> 2. I read a few times that ferrites in inductors (and I assume transformers as well) can be a real problem for 1/f noise. I'm using a transformer (TC1-1t from minicircuits) which is made with a ferrite bead. Do you know of any transformers or inductors that have low 1/f noise ferrite material. >>> 3. Have I missed something fundamental in the design. The goal is to build a very good oscillator. I would like to achieve something like -110dBc/Hz at 10Hz distance at 33.8688MHz. >>> >>> Thanks in advance for having a look and best regards, >>> >>> Hans Rosenberg >>> >>> >>> >>> >> 1) A LED plus a BJT and resistor makes a much lower noise current source >> than a zener based reference plus a BJT and resistor. >> Some shielding of the LED from ambient light may be required as LEDs (as >> all PN junctions) are photosensitive. >> A resistor in series with an RF choke can be used to replace the current >> source. >> >> 2) A BJT with a correctly proportioned coupling network will not >> significantly load the crystal. >> >> 3) There is little isolation between the oscillator and the load, a >> slight change in topology will improve this. >> >> 4) Modified Pierce overtone crystal oscillators using JFETs were popular >> in the 1960's and 1970's. >> >> 5) The amplitude limiting mechanism in the oscillator is important as it >> affects the phase noise. >> >> 6) A lower noise audio BJT will make a much lower (flicker) noise >> current source than a 9GHz transistor. >> >> 7) Its usually better to use the crystal to filter the output signal. >> >> 8) One of the Driscoll BJT oscillators or a variant thereof is a good >> stating point. >> AGC via a varactor based attenuator can be used. >> The output signal can be extracted from the second transistor collector >> using a transformer with little interaction with the collector tank tuning. >> >> >> Bruce >> >> >> > Addendum > > 9) Having a dc voltage across the crystal isnt a good idea as it can > lead to unwanted phase modulation. > > > 10) The current source transistor collector base voltage seems a little low. > > Bruce > > Addendum II 11) A common base cascade will have much lower noise than a cascode chain, however the CB cascade needs a current input. This can be derived in your modified Pierce circuit by connecting C1 to the emitter of the input CB transistor rather than to ground. The transformer is then not required. A lower impedance input at the CB stage emitter can be achieved by replacing the Cb transistor by a CECB feedback pair. Whats the limiter circuit? Is it at the cascode buffer chain output? Bruce _______________________________________________ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.

I would like to add some perspective to this discussion. The 10811 oscillator simply uses a plain vanilla Pierce circuit configured so that one terminal of the crystal is at ground. The base emitter capacitor is replaced by a mode suppression network to force operation to the correct overtone and mode (C as opposed to B). It uses a common (at the time) transistor, the 2N5179. The AGC varies the DC collector current in the transistor. A zener diode of a type which is known to have low noise is used to bias the varactor. There is no magic here. The results are that below 100 Hz, the noise is strictly determined by the intrinsic crystal noise (yes, crystals have their own noise). Intrinsic crystal noise can be measured with an HP3048 type phase noise system, if you know what you are doing. At 1 kHz and above, the noise is strictly determined by the buffer amplifier following the oscillator circuit. Read Rob Burgoon's patent to understand why his patented buffer is the best that can be built (ie limited by physics). The actual buffer used in the 10811 has some compromises in it that increase the phase noise floor. You can easily get back to optimum performance by using a string of common base amplifiers. It is therefore puzzling to me that there is a search for the holy grail low noise oscillator circuit. What is it that you do not like about the 10811 circuit? Rick Karlquist N6RK

Hans Hans Rosenberg wrote: > Hi Bruce, > > Thanks a lot for all the input. > > Do you have any figures on the noise voltage on a led? A good buried zener reference does 100nV/rt Hz, what does a led do? > > A good one (like all forward biased PN junctions) will be much quieter than a zener. > You mention that an amplitude control mechanism influences phase noise. Are you suggesting that a good amplitude control is better than turning off the active device for some time each cycle? > > I started having a close look at the driscoll design. It looks really nice. It seems much more elegant than what I was trying to do. I do have one question about the AGC circuit. The tune-voltage must be kept in the middle of the supply, otherwise the amplitude control loop will modulate the overall capacitance of the capacitive attenuator. So this is critical to set correctly in practice. There probably is a good reason why the current in the transistor is not changed to achieve amplitude control. Do the capacitances in the transistors vary so much that they start introducing more phase errors than when tuning a capacitive attenuator correctly? Or is there another reason. > > My isolation chain is indeed build with cb stages and not cascodes (wrong terminology from my side). I generate a current using a degenerated differential pair (the solution in the driscoll oscillator looks much more elegant). My limiterstage is build with a differential pair with 2 resistors in the collectors. The differential limited voltage goes to a transformer. The secondary side of that transformer is biased to 1.4V and drives a 74ABT inverter. I choose these inverters for their very low output impedance so I can drive a load with a high slewrate. However too much slewrate may cause a lot of interference and maybe reflections, so I'm thinking about putting a 33OHm resistor in series with the output to limit the bandwidth and reflections somewhat. Do you have any experience which family of ports (ac, hc, abt etc) delivers the best phase noise results (I should not have that much problems since I'm driving it with a pretty good squarewave but am curious what you think of it). > > A self biased inverter with resistive feedback between input and output driven directly with the RF signal (~1V rms amplitude for 3V Vcc) usually has a slightly lower phase noise. However one then needs to ensure that the RF input signal is always present (otherwise Vcc current can be high). RC decoupling the Vcc supply of the inverter is also useful. I have no data on ABT devices, you would need to measure it - not too difficult (no phase lock loop required just a phase shifter) with a splitter, sound card, a mixer etc. > I also would really like to get my hands on a 33.8688MHz SC-cut crystal, but I found they are hard to get. I guess this means starting up a production run at a manufacturer which is quite expensive. And I guess I'd have to include a temperature control circuit since the temperature characteristics of SC-cut crystals are not that good. > > Using a LED within an oven may not be that reliable, depending on the LED packaging and operating current. > Nice to hear that you're into optics as well! I build some basic telescopes, although I never made my own lenses or mirrors. But I'm very fascinated by the subject. I used to work for ASTRON, which is a company that builds radiotelescopes. I also used to give stargazing evenings for tourists at a local observatory which was a lot of fun. We had a 60cm mirrortelescope there. Unfortunately, this is a bit of overkill in holland with the bad seeing we usually have here. > > The local astronomical society has a 61cm classical Cassegrain plus a 14"Meade and various Dobsonians. We also have a dish for radio astronomy (but no hydrogen maser). > Thanks again for the nice input. > > Best regards, > > Hans Rosenberg > > ________________________________________ > From: time-nuts-bounces@febo.com [time-nuts-bounces@febo.com] On Behalf Of Bruce Griffiths [bruce.griffiths@xtra.co.nz] > Sent: Wednesday, January 14, 2009 10:41 PM > To: Discussion of precise time and frequency measurement > Subject: Re: [time-nuts] Ultra low noise Pierce oscillator??? > > Bruce Griffiths wrote: > >> Bruce Griffiths wrote: >> >> >>> Hans >>> >>> Hans Rosenberg wrote: >>> >>> >>> >>>> Hi Time-nuts people, >>>> >>>> I'm new to this forum and really enjoyed reading all the design strategies on oscillators. >>>> >>>> I'm currently designing a low noise oscillator and I wonder if you can help me with some questions. I've attached a picture of the oscillator core I'm using. I decided to go for a Pierce design some time ago after looking at a few topologies. I took a number of measures to give the design a low 1/f noise. >>>> >>>> >>>> 1. Low DC gain. There is a transformer in the drain line. >>>> 2. Ultra low noise supply (not on the schematic) using an ad586 buried zener reference with a -3dB lowpass at 0.1Hz. This 5V reference is then increased to 12V using a low noise discrete buffer. >>>> 3. The phase shift across T1 is only a few degrees (<5degrees), low phaseshift in the amplifier reduces 1/f noise. >>>> 4. I chose a fet in order to minimize the load at the 'output' of the crystal, the only load I have now is the bias resistor of 100k. This resistor does cause low frequency noise (at higher frequencies C1 shorts it). This may be the problem in this circuit, Cgd is modulated by this noise and a low frequency voltage is applied at the gate, which is however not amplified because of the currentsource in the source line which should cause nearly infinite feedback for low frequencies making the DC gain even lower. >>>> 5. The current source produces low frequency noise. I have to have a current source though (I could use a resistor but I calculated that produces more noise). I could increase the voltage across R5 and make R5 bigger to reduce the noise, however, this will reduce Vgd which means modulation of Cgd becomes worse. I think I'm ok here by dividing the voltage across the currentsource and the active oscillator element in half. Cgd of a J309 is around 2.5pF at Vds=10V (it is a little higher in my case since Vgd is lower) >>>> 6. L2 can be mounted to accommodate overtone crystals. ( I still have to calculate a value for L2 and C2 to get the correct impedance at resonance) >>>> 7. F1 and F2 are ferrite beads with a low impedance at DC (far less then an ohm) and rising impedance at higher frequencies to prevent oscillation of the RF transistors. >>>> 8. I've found a really good overtonecrystal from Citizen. CM309S. I measured the unloaded Q to be 313000, the loaded Q (using simple estimation) should be around 280000 in the circuit. (C0=2.5pF, R=20Ohm, C1=0.75fF, L=29.444mH, can be aquired at digikey). The spec for Rs was <130Ohms, I guess reality is much better :-) >>>> 9. The transformeroutputs are going to an isolationchain using 3 cascodes and then a discrete limiter not shown on the schematic. >>>> 10. This whole oscillator core will be placed in an RF shielding can. The isolationchain will have a can of it's own to so pulling should be nearly eliminated. >>>> >>>> Now my questions are: >>>> >>>> >>>> 1. I don't see a pierce design in any of the low noise oscillator circuits in the discussion threads about low noise oscillator design. Is there something fundamentally wrong about this topology? >>>> 2. I read a few times that ferrites in inductors (and I assume transformers as well) can be a real problem for 1/f noise. I'm using a transformer (TC1-1t from minicircuits) which is made with a ferrite bead. Do you know of any transformers or inductors that have low 1/f noise ferrite material. >>>> 3. Have I missed something fundamental in the design. The goal is to build a very good oscillator. I would like to achieve something like -110dBc/Hz at 10Hz distance at 33.8688MHz. >>>> >>>> Thanks in advance for having a look and best regards, >>>> >>>> Hans Rosenberg >>>> >>>> >>>> >>>> >>>> >>> 1) A LED plus a BJT and resistor makes a much lower noise current source >>> than a zener based reference plus a BJT and resistor. >>> Some shielding of the LED from ambient light may be required as LEDs (as >>> all PN junctions) are photosensitive. >>> A resistor in series with an RF choke can be used to replace the current >>> source. >>> >>> 2) A BJT with a correctly proportioned coupling network will not >>> significantly load the crystal. >>> >>> 3) There is little isolation between the oscillator and the load, a >>> slight change in topology will improve this. >>> >>> 4) Modified Pierce overtone crystal oscillators using JFETs were popular >>> in the 1960's and 1970's. >>> >>> 5) The amplitude limiting mechanism in the oscillator is important as it >>> affects the phase noise. >>> >>> 6) A lower noise audio BJT will make a much lower (flicker) noise >>> current source than a 9GHz transistor. >>> >>> 7) Its usually better to use the crystal to filter the output signal. >>> >>> 8) One of the Driscoll BJT oscillators or a variant thereof is a good >>> stating point. >>> AGC via a varactor based attenuator can be used. >>> The output signal can be extracted from the second transistor collector >>> using a transformer with little interaction with the collector tank tuning. >>> >>> >>> Bruce >>> >>> >>> >>> >> Addendum >> >> 9) Having a dc voltage across the crystal isnt a good idea as it can >> lead to unwanted phase modulation. >> >> >> 10) The current source transistor collector base voltage seems a little low. >> >> Bruce >> >> >> > Addendum II > > 11) A common base cascade will have much lower noise than a cascode > chain, however the CB cascade needs a current input. > This can be derived in your modified Pierce circuit by connecting C1 to > the emitter of the input CB transistor rather than to ground. > The transformer is then not required. > A lower impedance input at the CB stage emitter can be achieved by > replacing the Cb transistor by a CECB feedback pair. > > Whats the limiter circuit? > Is it at the cascode buffer chain output? > > Bruce > > Bruce

Hans Hans Rosenberg wrote: > Hi Bruce, > > Thanks a lot for all the input. > > Do you have any figures on the noise voltage on a led? A good buried zener reference does 100nV/rt Hz, what does a led do? > > You mention that an amplitude control mechanism influences phase noise. Are you suggesting that a good amplitude control is better than turning off the active device for some time each cycle? > > There is some debate about this in the literature. > I started having a close look at the driscoll design. It looks really nice. It seems much more elegant than what I was trying to do. I do have one question about the AGC circuit. The tune-voltage must be kept in the middle of the supply, otherwise the amplitude control loop will modulate the overall capacitance of the capacitive attenuator. The input capacitance doesn't vary that much if the attenuator is correctly proportioned: http://www.karlquist.com/97bri.pdf > So this is critical to set correctly in practice. There probably is a good reason why the current in the transistor is not changed to achieve amplitude control. Do the capacitances in the transistors vary so much that they start introducing more phase errors than when tuning a capacitive attenuator correctly? Or is there another reason. > > > Thanks again for the nice input. > > Best regards, > > Hans Rosenberg > > Bruce