Several people pursued the Janascard oscillator and were not able to get the same performance. A single frequency oscillator is an easier target and the best and most cost effective right now is Victors. He says he gets -140 to -150 harmonics which is pretty much all you can ever expect. I'm waiting for the new 4 oscillator thing from Germany. Its a little bit of an odd duck but if it performs, great.
That cap is there to take the high frequency peaking out. It's in the the local feedback and doesn't effect the loop that much. I have such a cap in my oscillator and have not seen any adverse effect from it being there. The phase lead capacitor discussed works independently.
I've also used a small 10 ohm resistor in series with the highest range caps. The LT1468 needs a small FB cap to null the input capacitance. This is explained in the data sheet.
because of this too much resistance in series with the integrator caps will cause rf oscillation so a phase lead cap is better. Each op amp kind will have it's own requirements.
I've also used a small 10 ohm resistor in series with the highest range caps. The LT1468 needs a small FB cap to null the input capacitance. This is explained in the data sheet.
because of this too much resistance in series with the integrator caps will cause rf oscillation so a phase lead cap is better. Each op amp kind will have it's own requirements.
Surely the gain in the outer loop is simply the product of the gains of each individual section.
So if we increase the local -ve feedback to reduce the inverter gain then it must reduce the overall loop gain by exactly as much?
I don't see how it can be independent, as stated above.
It's not like the local feedback is in a different frequency band to the phase lead capacitor in the overall loop.
In fact GK's local loop capacitor reduces the over loop gain about an octave before the 505 tries to increase it.
Different op-amps so not directly comparable, but fairly similar frequencies.
I'll think about that, in the meantime, why the LT1468?
Optimized AD797 looks to have excellent distortion performance and about 10 dB less noise.
That's a substantial improvement in an ultra-low distortion oscillator where noise really matters, I believe.
Was it just the price of the AD797 or concern about stability?
Best wishes
David
Last edited:
In a re-do of my oscillator, my choice for the op amp would probably be the LM4562. It has lower voltage noise than the 5532, but higher input bias current, implying higher input current noise. The chosen impedances for the R and C tuning components must be carefully evaluated for noise in conjunction with the integrator op amps. In my oscillator, I would probably go to at least 2X the current capacitance with a corresponding reduction in the Rs.
Cheers,
Bob
The current noise is the essence when choosing a op amp for an oscillator. It seems the the lower voltage noise op amps have higher current noise. The overall noise might be lower using am op amp with higher V noise and lower current noise considering the tuning resistors.
I paralleled two Mdacs in the integrator sections to get the tuning R down and double the C.
I paralleled two Mdacs in the integrator sections to get the tuning R down and double the C.
Hi Dave,
Let's stay within the context here. The Fb cap is to take out high frequency peaking.
The peaking does nothing for bandwidth but does causes instability (high frequency oscillation). With the LT1468 this peak is about 7Mhz. We add just enough Cfb to reduce or flatten the peak out, balance the input C to form a null. If Cfb is chosen correctly it does nothing to the bandwidth. Of course if we go nuts and add 1nF cap then yes. The Cfb supplies the current to the input C so the input circuitry doesn't, effectively canceling a pole.
The Cfb can only reduce gain to unity. It's not the same as internal compensation which keeps on going well below unity. There is BB app note which explain this. I can dig it up if you like.
Let's stay within the context here. The Fb cap is to take out high frequency peaking.
The peaking does nothing for bandwidth but does causes instability (high frequency oscillation). With the LT1468 this peak is about 7Mhz. We add just enough Cfb to reduce or flatten the peak out, balance the input C to form a null. If Cfb is chosen correctly it does nothing to the bandwidth. Of course if we go nuts and add 1nF cap then yes. The Cfb supplies the current to the input C so the input circuitry doesn't, effectively canceling a pole.
The Cfb can only reduce gain to unity. It's not the same as internal compensation which keeps on going well below unity. There is BB app note which explain this. I can dig it up if you like.
LM4562 has typically input bias current 10nA, but NJM5532 has typically - 200nA.
Many examples of LM4562 (and the same chips LME49720) have the low frequency noise problems, and must be selected before using in an high performance applications:
https://e2e.ti.com/support/amplifiers/audio_amplifiers/f/6/t/415907
Best regards,
Victor.
My understanding was that the LM4562 has not been discontinued. Are you sure?
Cheers,
Bob
I thought I saw it on the list but I will look again if I can find it.
LME49720 has exactly the same data sheet specs and thought to be the same IC.
If it goes...
I do not have the datasheets in front of me, but I seem to recall that the LM4562 input bias current is higher than that of the 5534. I think the relationship is pretty much the same for the 5532.
I have not seen a serious 1/f or popcorn noise problem in the 4562, but all of the ones I have used were bought at the same time. Perhaps others here can comment on their experience with LF noise in these devices.they seemed to be saying that only about 10% were significantly bad in 1/f noise, but they did say that screening by the user was needed if best noise performance at LF is to be obtained. For an oscillator expected to perform to -120 or so down to 20Hz, I think it is likely that the builder might want to try a few samples in each location anyway. Unfortunately, this is probably not practical for SMT implementations.
On the discussion you linked to,
Cheers,
Bob
It's on the list but may be restricted to package.
http://www.proaudiodesignforum.com/images/pdf/ti_20150901004_09292015_eol.pdf
According the data sheets input bias current for these chips are that:
LM4562 - 10nA http://www.ti.com/lit/ds/symlink/lm4562.pdf
NJM5532 - 200nA http://www.njr.com/semicon/PDF/NJM5532_E.pdf
NJM5534 - 500nA http://www.njr.com/semicon/PDF/NJM5534_E.pdf
but I did not measure them in practice...
In my practice around 50% of the LM4562 indicate very good noise performance, but around 20% goes to the trash. This job for to select is really painful. Noise may be unstable in time - within some seconds may be OK, but later may go to the critical level.
Best regards,
Victor.
The LME47720 is the same die as the LM4562. They renumbered for the new family. It was a marketing decision. The whole family is doomed. TI is shutting down the fab and I was told that the process was not portable to another fab. TI has similar parts so they don't see a need to keep them.
The input bias number will be misleading since these chips will have input bias cancellation stuff. You really need to measure the noise. I was told that the internal leakage currents are the best indicators for selecting the best chips, and that's why there are so may variants of the same thing in the LME family. I'm not sure how you would measure that on the packaged chip.
The input bias number will be misleading since these chips will have input bias cancellation stuff. You really need to measure the noise. I was told that the internal leakage currents are the best indicators for selecting the best chips, and that's why there are so may variants of the same thing in the LME family. I'm not sure how you would measure that on the packaged chip.
Probably you are right about the input bias current cancellation in the LM4562. The offset current and the input impedance is similar as for the NJM5532, and maybe the "real" input bias current is also similar.
Best regards,
Victor.
As far as I can see the LME49720 does survive in two versions, the LME49720MAX/NOPB (SO8) and the LME49729HA/NOPB (TO-99, metal can).
See LME49720 | Audio Operational Amplifier | Operational Amplifier (Op Amp) | Sample & buy
See LME49720 | Audio Operational Amplifier | Operational Amplifier (Op Amp) | Sample & buy
I'll see if i can get a straight answer from my TI FAE.
Bruce Hofer discusses composite opamps in his latest newsletter, mentioning separate supplies for two opamps and hinting at some other challenges. He mentions that these techniques were used in the AP555x. Anyone else with experience? I have built composites with opamp and LM3886 that worked quite well, but that was 25 years ago. The composite in the Linear tech 10 KHz oscillator is notoriously fussy.
Bruce Hofer discusses composite opamps in his latest newsletter, mentioning separate supplies for two opamps and hinting at some other challenges. He mentions that these techniques were used in the AP555x. Anyone else with experience? I have built composites with opamp and LM3886 that worked quite well, but that was 25 years ago. The composite in the Linear tech 10 KHz oscillator is notoriously fussy.