Only if the noise temperature of the source is 290 K. A dynamic microphone at room temperature in total silence would be quite close, but line level sources normally have far more noise than the thermal noise of their output impedances.
-113 dBu is about 1.7 uV RMS. If the bandwidth is 20 kHz, that's some 12 nV/√Hz, roughly the thermal noise of 9 kohm at 290 K. Line level sources usually have output impedances far below 9 kohm.
I meant the power supply.
You could try putting some brute-force low-pass filter between the power supply and the circuit, 100 ohm and 1000 uF electrolytic cap or so, to see if that does anything. Or check if the noise spectrum at the output changes much with the source replaced with a short - just to rule out that out-of-band noise from the source has anything to do with it.
Last edited:
This statement is beyond a little for me. I am cold on this. How should I understand it? Can you please try to rephrase, maybe I'll catch more of your idea?
So, what can be inferred from -113dBu noise floor at Input?
And, what then can be inferred from -103dB noise floor on the amp Out?
Noise figures are only easy to interpret when the source is some passive thing close to room temperature that produces only thermal noise. In most other cases, it is easier to just specify noise levels and bandwidths, or noise densities.
-103 dBu in 20 kHz at the output is more than I would expect, hence the proposals for extra experiments I added to post #43, just to see whether the noise really originates from your circuit.
-103 dBu in 20 kHz at the output is more than I would expect, hence the proposals for extra experiments I added to post #43, just to see whether the noise really originates from your circuit.
I see a diode, a Ker and an additional Elko.
I understand only the Ker. Can you please describe the others / update schematic?
Note, these numbers I gave are the SA noise floor results (densities & averaged). Not the total noise in 20KHz BW.
The SA display is set to 20KHz: but behind this there is no filter. I will see mirrors of signals and noise above the display range.
Maybe the extra noise comes from "out of display band"... theoretically possible.
But, the 0-1000Hz still shows same "slope" feature: if the out-of-display-band noise would be so big (accounting for 10dB), then the noise floor at 0-1KHz will loose most of its slope - in my expectations. Right?
Additional tests: (all suggested above by your kind help):
- will try to filter more the PS (not much hope as I did so already)
- will try to implement the PS+"denoisator"
Cool stuff. - will put Schottky diodes and pre- + after- filters in the PS, before testing "the denoisator". Some parts needs order.
- will buy a ZTX1015A, just for tests (needs order)
- will put the FET CC instead of R4. (will this not increase the OUT noise?)
- I removed already the R1, but did not tested it without it.
- will try on the breadboard the LM137 thing: focus for SNR.
- will do more checks of SA: also will do screenshots to share here
- additionally, I can temporarily add LP 20KHz 1-pole filters (for Source, for Amp and for Probes). Would this be enough?
- last but not least, I can test with a Lab PS (more available).
Good things list for this problem:
- the -103dB seems also unexpected for you as well
- you did not found so far anything essentially wrong in my protocol.
Last edited:
Sorry, this still needs more clarification. My current understanding is maybe false, that NF is especially useful for active circuits.
Any hint for a reference-level book?
It will help me also at work, to understand NF in more complex systems having various parts at various temperatures.
Maybe a silly question:
Look in my circuit: it has on the line Out a divider as load, which is made of a 560 and a 9K resistor. Maybe this is the problem?
Adding to the list of to-do tests:
- cold spray for 9K on Line Out divider.
Here it is:
What you identified as a diode is simply the 10 ohm carbon composition resistor
Thanks LV
I thought it is a reverse protection diode cross the IN/Out of the regulator, but C1 is not such a big reactive load to put any problems. Besides then, the Keramik missed a resistor to make a snubber. Sorry it just looked very similar to a diode !
I wanted yesterday to ask you at which frequency the snubber should be. Now I know
I thought it is a reverse protection diode cross the IN/Out of the regulator, but C1 is not such a big reactive load to put any problems. Besides then, the Keramik missed a resistor to make a snubber. Sorry it just looked very similar to a diode !
I wanted yesterday to ask you at which frequency the snubber should be. Now I know
I ought to add here a few more entries:
- there is a plan of debug
- there is a very nice help from experts
LV, can you please comment?
Clearly, the distortions will be lower compared to one-BJT + R4.
But Noise? What behavior should be expected?
Don't worry. A JFET current source usually produces more noise than a resistor with the same DC current and DC voltage drop, but that noise will still be negligible compared to the collector shot noise of the bipolar transistor.
By the way, parasitic oscillations could be a cause of excessive noise, as could a damaged transistor.
By the way, parasitic oscillations could be a cause of excessive noise, as could a damaged transistor.
I have measured the THD of the 317 circuit (@ 1kHz, 8Vpp out, 560ohm load): it is 0.0015%, thus a smidgeon above LTspice's prediction, as expected.
In the same conditions, the full power bandwidth is ~36kHz. Not extravagant, but certainly sufficient for regular audio applications
In the same conditions, the full power bandwidth is ~36kHz. Not extravagant, but certainly sufficient for regular audio applications
An afterthought: the 36kHz limit is most probably caused by the reactance of the Zobel's capacitor: 10nF @36kHz is 445ohm, to be compared with the 470ohm current sink.
Lowering the 470ohm would probably extend the bandwidth, at the expense of the current consumption; however it would also increase the available output power.
BTW, here is the .asc if someone wants to play with it:
Lowering the 470ohm would probably extend the bandwidth, at the expense of the current consumption; however it would also increase the available output power.
BTW, here is the .asc if someone wants to play with it:
Yes, it seems lowering R3 to 100 ohm makes it suitable to use 80 Ohm headphones but it needs cooling.
I have tweaked the circuit, and remeasured its performance. The optimum value for the Zobel's resistor is 4.7ohm, but 10nF for the cap remains optimal: going lower endangers the stability:
I have remeasured the THD in better conditions (better GND connections, etc.) and it is in fact half of what I measured initially: 0.0007%. Note that the 317 is a cheap Chinese one, bought from Banggood
I have remeasured the THD in better conditions (better GND connections, etc.) and it is in fact half of what I measured initially: 0.0007%. Note that the 317 is a cheap Chinese one, bought from Banggood
This is great news!
Measuring THDs so low, means to me that I expect SNR is also equally low. Am I right?
How comes that few years ago, others measured SNR for 317 in the range no better than 80dB (after improvements!) ? If I may ask, what is changed here? For example, how much better must the GND be? A new picture is welcome. Anyway, the previous setup was equally impressive.
I go in the lab right away.