The noise plots seem to show that Syn08's circuit and yours have almost the same noise performance. But it is not as it looks like. The RTI noise levels at 100R source resistance are a little bit lower as yours (2.11 nV/rtHz vs 2.38 nV/rtHz) but the gain is quite different: at 100 R Syn08's circuit has a gain of x0.6! while your's is at x9.4.
For gains lower than ~x3, all circuit elements contribute to output noise while for gains > 3 the noise contribution of all elements except the input transistors and emitter resistors gets neglegible.
When increasing the gain of Syn08's circuit (increase 150R resistors and increase supply voltage to keep currents in the input BJTs constant) the picture is quite different:
at Rs = 10R: RTI noise is 498pV/rtHz vs 543pV/rtHz (10R alone is 404 nV/rtHz)
at Rs = 30R: RTI noise is 853pV/rtHz vs 1.02nV/rtHz (30R alone is 703pv/rtHz)
at Rs = 100R: RTI noise is 1.55nV/rtHz vs 2.38nV/rtHz (100R alone is 1.28 nV/rtHz)
So noise figure is going slightly down with Syn08's circuit as expected from an amp with low input noise current (from 1.77dB at Rs = 10R to 1.66 dB at Rs = 100R) while for your amp it is significantly going up (from 2.5dB at Rs = 10R to 5.38dB at Rs = 100R)
When using current mirror deg. resistors of 11R as per your first schematic schown instead of the 91R as used in the last simulation, the increase in noise figure gets significantly more worse.
Last edited:
Here is the paper
Noise in Current Sources and Current Mirrors
(If there are any objections to me hosting it, pm me and I will take it down)
Noise in Current Sources and Current Mirrors
(If there are any objections to me hosting it, pm me and I will take it down)
What is interesting is not the fact that current mirror noise goes down if degeneration resistors are made larger and vice versa but the fact that the current mirror noise is not simply added to all other noise sources but is dependant on the input source resistance and seems to appear at the amp input.
Taking the amp noise with Rsource = 0 as the input voltage noise of the amp and doing some math for calculating back the input current noise from simulated RTI noise voltage and subtracting the amp input voltage noise and the source resistor noise, you get an input noise current density of 85pA/rtHz for Rs = 3R going down to 36pA/rtHz for Rs = 10R, 25pA/rtHz for Rs = 30R and asympotically approching 20 pA/rtHz for higher Rs.
So we have some gain function for the observed exess noise depending on the source resistance despite the fact that the current mirror is directly connected to the output and via the base blocking capacitors at the input BJTs to signal ground.
I skimmed through the paper quickly. The authors have modeled all the noise sources as inputs to a perfect common emitter amplifier. The noise is therefor approx amplified by the collector load divided by re’. When you degenerate the current mirror, the gain approximates to Rl/(re’+RE) - ie it’s much lower so the input referred noise is not amplified as much.
I have not quite worked out why it’s asymptotic, but presume what’s happening is that below a certain gain RE noise and input referred noise dominate, so no further reductions in noise are available by reducing the gain further through increasing RE.
I have not quite worked out why it’s asymptotic, but presume what’s happening is that below a certain gain RE noise and input referred noise dominate, so no further reductions in noise are available by reducing the gain further through increasing RE.
Last edited:
I'm glad to see that doing some math decrease the input current noise from 58nV/rtHz to 20pA/rtHz.
I have not the time to make simulation and calculation right now but I think this behaviour is only due to the current mirror noise into the output impedance of the amplifier.
This can be lowered by increasing the degeneration resistors and the power supply.
At first I thought Herve had converted the excellent Wonky Virtual Earth performance of pure Duraglit & syn08's #375 back to a common or garden Common Base with its max 3dB NF .. but closer examination of #1297 shows a simpler explanation.
Using the nomenclature in #1297 ..Q5 & 6 are a Duraglit but with extra EVIL resistors adding noise & introducing THD.
Q7 & 8 convert that to syn08's Lesser Leach which has only slightly worse performance
In syn08's #375, this biasing chain is fed by R4 & R5, 150R which simply act as part of the load in parallel with R7, the 1K load resistor.Herve perhaps thought that replacing these resistors with current mirrors would improve performance but they don't.
His Q10, 12 & J1 chain is no problem. It's directly across the supply so, to a first approximation, has 'no' connection to the output. The current in Q10 is also the current in Q12 (less the small base currents).
That's why pure Duraglit has very OC low noise too.
But Q9 & 11 are two uncorrelated current sources so their noise is fed directly to the output. Bigger R11 & 18 make them less EVIL. Bigger source resistance R13/Rg makes their EVIL contribution larger compared to signal.
(rant) .. why would you use 4.5x the number of active devices, zillion more passives, 3x the current bla bla ... to get worse performance ...
(/rant)
Last edited:
The 58nV were my error, I made pV to nV
Just could also confirm with a couple of sims that it is indeed the current mirror noise current flowing through the output resistor creating an output noise voltage. The relative change to amplified input noise was simply the fact that gain changes with these amps with changing source resistor.
What is the purpose of the current mirror? What benefits does it bring compared to syn08's #375 or true Duraglit?
In particular, what extra benefits does it give to a user or builder?
@syn08
That suggestion to have my PCB’s made by JLCPCB was very welcome, thanks.
They charge a fraction of what I used to pay locally and their return time is second to none.
Anxious looking forward to publish noise figures on my Inverting and Non Inverting universal Head Amp based on these ZTX trannies.
Hans
That suggestion to have my PCB’s made by JLCPCB was very welcome, thanks.
They charge a fraction of what I used to pay locally and their return time is second to none.
Anxious looking forward to publish noise figures on my Inverting and Non Inverting universal Head Amp based on these ZTX trannies.
Hans
The 58nV were my error, I made pV to nV
Just could also confirm with a couple of sims that it is indeed the current mirror noise current flowing through the output resistor creating an output noise voltage. The relative change to amplified input noise was simply the fact that gain changes with these amps with changing source resistor.
That's correct.
Increasing the power supply to 6V and the degeneration resistors to 400 ohm, give an input current noise density of ~10.5 pA/rtHz and an input voltage noise density of ~220pV/rtHz.
The purpose to the current mirror was to increase the gain of #375 schematic with ~same noise.
The gain is increased but at the expense of an input current noise density penalty caused by the current mirror noise.
So the schematic could be competitive for very low resistive cart, but have absolutly no interest for the others.
Just uploaded today, really yesterday, 2 boards to PCBway.
One of my boards was bigger than 100*100 mm by 6 mm, that exploded it's price by *5.5.
Still OK. First time they get a little more than DHL for the transport.
It was Sunday, also in China, but they reviewed my boards immediately.
Next time I'll watch the form factor more closely.
I also ordered some stuff from DK, amongst it the npn and pnp ZTX trannies.
Somehow I always find interesting stuff until it hurts in the wallet.
Nevertheless I included some test structures, a "50 Ohm" line for the TDR,
Some test-SMAs for the VNA and some GaN transistors. That will stress their solder mask, and maybe more.
That are just nekkid chips with 4 bumps for reflow solder.
< EPC2038 EPC | Diskrete Halbleiterprodukte | DigiKey >
The bottom 2/3 are just another FET preamp.
Some test-SMAs for the VNA and some GaN transistors. That will stress their solder mask, and maybe more.
That are just nekkid chips with 4 bumps for reflow solder.
< EPC2038 EPC | Diskrete Halbleiterprodukte | DigiKey >
The bottom 2/3 are just another FET preamp.
Attachments
Last edited:
If you are using your circuit for a MC head amp, your maximum gain is going to be limited by the 47K i/p of your MM RIAA amp.
#375's gain is limited by R4 & 5, which feed the current mirror and appear in parallel with the output resistor R7.
True Duraglit uses much larger resistors so can be matched to much higher gain. It also has slightly less noise & THD than #375 at half the current bla bla ..
If this isn't just a SPICE world exercise but you are building a real life device, pay attention to Bonsai & syn08's caveats about power supplies & solar cells or use batteries.
Nope. #375 gain is limited by the power supply voltage. It is independent on R4/R5.