That should be ..
"You just need to get the source resistance seen by your 'FET OPA' (via the transformer) sufficiently LARGER than its Rnv"
Mea maxima culpa
Richard, the design issues of the Lepaisant circuit have little or nothing to do with SUT's. As you note they get 60dB gain with 100kBW. I don't have the energy to redo their numbers for a 3 Ohm source since I'm assuming the low frequency corner would suffer, but it is not clear to me without doing the exercise that the noise figure for 3 Ohms would not be <1dB. I might be wrong, that's OK too.
SUT's are more complicated since the bandwidth depends on loading and I apologize for not noticing this, but it's your expertise and we are simply asking where the numbers come from because maximum power transfer has nothing to do with it.
Here's a chance to clarify, please take this Jensen applications note and explain where the 1.5dB NF comes from.
https://www.jensen-transformers.com/wp-content/uploads/2014/08/jt-115K-e1.pdf
What you have is a transformer bobbin that you fill with Cu wire. A given LF response needs a certain Inductance so a certain number of turns. But more turns means more series resistance so noise suffers. That's the LF/noise compromise but there's also an overload constraint at LF as less turns means the core saturates earlier. Often this is what you design to as 'inductance' of a 'metal' cored item varies with level but the overload level(s) is always simple to predict.
You design LN transformers by sitting down with the core & bobbin spec and a set of wire tables.
It's way past my bedtime but I'll have a look at the Jensen note tomorrow
... and yes. Maximum. power transfer has nothing to do with it
Last edited:
https://www.pearl-hifi.com/06_Lit_A...f_Low-noise_Amps/Design_of_Low-noise_Amps.pdf
Get it while you can.
Get it while you can.
Paper also contains a short paragraph about noise in photo detectors:
Photoconductive detectors produce generation-recombina*tion (GR) noise in response to a steady irradiance. Hole-elec*tron pairs are generated randomly and recombine randomly by a statistically unrelated process. Thus full GR noise ne*glecting that which is thermally generated. corresponds to twice shot noise on the absorbed background photon rate for intrinsic photoconductors. As the dc bias is increased, a volt*age is reached at which the minority carrier (hole) transit time is less than the lifetime. At this bias, the carriers are swept out of the device before they recombine and the GR noise ap*proaches shot noise on the photocurrent
This confirms Syn08's and my assumption that G-R noise is the main source for photo cells and that only that amount of light irradiation should be used to just deliver the required current. Then G-R noise gets minimal and only shot noise is left.
#1 A preamp with Rs=12ohm G=17dB and output noise divided by gain of 0.61nV/rtHz. Then your "broadcasting" definition of noise figure renders NF=20LOG(0.61/0.44)=2.8dB.
#2 A preamp with Rs=12ohm G=24dB and output noise divided by gain of 0.62nV/rtHz. Then your "broadcasting" definition of noise figure renders NF=20LOG(0.62/0.44)=2.8dB. Same thing.
So they essentially have the same "broadcasting" NF, which one is better for an audio MC application?
N.B. numbers are real, from two different versions of #375.
... noisewise they are identical. They have (almost) identical rti noise specs of 0.61/0.62 nV/rtHz. The better one is the one that fits your gain requirements best.
True. However, I re-measured for Rs=100ohm (exaggerated on purpose) and now the "broadcasting NF metric no longer converges.
#1 A preamp with Rs=100ohm (1.28nV/rtHz) G=1.23dB and output noise divided by gain of 2nV/rtHz. Then your "broadcasting" definition of noise figure renders NF=20LOG(2/1.28)=3.9dB.
#2 A preamp with Rs=100ohm (1.28nV/rtHz) G=24.0dB and output noise divided by gain of 1.77nV/rtHz. Then your "broadcasting" definition of noise figure renders NF=20LOG(1.77/1.28)=2.8dB.
Reason of the divergence is of course the higher current noise in #1 which multiplied by the higher Rs leads to a higher output noise.
The point is, this "broadcasting" NF metric is by no means more relevant than the standard voltage and current input referred noise. The voltage and current input referred noise have the net advantage of fully characterizing the LNA, independent from any external conditions, no need to specify a reference Rs. That's why they are widely used in EE, from system design, to devices data sheets.
Yes, thanks.
There are other papers that may be of interest on the Pearl site such as Battery Noise Measurements . In fact their whole literature archive is worth a look for the audiophile - see here.
All absolutely true. NF metric is a different description of absolutely the same fact and not more or less relevant. The difference in NF of 1.1 dB in your example is nothing else than the ratio of 1.77nV/rtHz rti and 2 nV/rtHz rti in dB
But NF is a relative measurement referenced to a system impedance while input ref. current and voltage are absolute not needing any reference. And this is the reason why NF is not popular anymore.
Last edited:
In RF it is still common and then there is noise temperature just to add to the confusion. Radio astronomers are fond of that. Ulrich Rohde has written some great stuff on noise in communications systems including some match terminated RF amps with very low noise figure that were published in ham radio magazines of all places.
forgot to add ... not popular anymore for audio. In the days when all recording and broadcast studios were fully analog, NF etc were the standard way for specifications. And audio engineers were using power as a reference. 1mW of power at a 600R impedance was used to describe signal levels (dBu, 0dBu is the mystery 0.775 Vrms level and +6dBu is the standard line level in professional studio equipment). And such specs are still used by e.g. the audio xfrm people. THD for signal transformers is still specified for dBu input levels.
More than once is referred to trying to prevent too large currents flowing through the Cart at switch on.
Just working on my inverting and non-inverting "Phantom" design, I noticed that Cart currents at switch-on are not the thing to watch but much more the switch-off currents.
Even the Duraglit, having exceptionally low switch-on currents in the nA range, shows a switch-off current of around 100uA, see image below.
Although the discharge waveform may differ, this current max is almost independent of Rcart and the size of the caps.
This seems to confirm "rumours" that short lasting currents in the hundreds of uA are causing no problem to an MC Cart.
Hans
Just working on my inverting and non-inverting "Phantom" design, I noticed that Cart currents at switch-on are not the thing to watch but much more the switch-off currents.
Even the Duraglit, having exceptionally low switch-on currents in the nA range, shows a switch-off current of around 100uA, see image below.
Although the discharge waveform may differ, this current max is almost independent of Rcart and the size of the caps.
This seems to confirm "rumours" that short lasting currents in the hundreds of uA are causing no problem to an MC Cart.
Hans
Attachments
Last edited:
Currents in the 100 µA range should be no problem. When running the cartridge in 'current mode', such currents could be the normal operating condition: e.g. many Ortofon MC cartridges have source resistances in the 5R range and output voltages of 500µV (5cm/s @ 1kHz). This will result in 100µA rms signal current.
That's with the cartridge in short; for the commonly recommended input impedance (100ohm) the current is much smaller.
Not saying that it matters, anyway, only that there's a large population out there that will freak about 100uA current through the cartridge.
The input impedance of the Duraglit (and #375 as well) is about 1/2gm that is (for Ic=2.5mA) about 5ohm so yes, it is reading a regular MC (Rs=10...20ohm) almost in short.
Thanks for this Gerhad.
It appears NF is used by more than Broadcast Organisations and old fogey Broadcast mixing desk people
Also da academics & those who's bedtime reading is IEEE Trans & Procs It's a good & accurate explanation of En, In & how they affect the noise performance for various sources .. though I would hardly call it 'quite simple & straightforward'. GG Baxandall's 1967 article is simpler and his last words on noise even clearer for us unwashed masses.
Netzer's eqn below his eqn (10) is what I pre10ed to have forgotten and his Fig 9 shows the range of good LN performance with NF at Ropt I was referring to. As Guru Wurcer points out, NF at Ropt is very good and this range is very large for LN FET i/p amps cos the very small In. But to take advantage of this performance, you need HiZ windings on your transformer w/o HF nasties.
The real reason NF is not so popular these days is cos most 'high quality' mikes are condensor and hence 'smallest En' is an appropriate LN metric for their mike preamps.
NF is still a useful concept when dealing with Electrodynamic Transducers, an important example of which is MC cartridges
Here, NF tells you how close you are to 'perfection @ NF = 0dB'.IMHO, only the BBC 4038 (considered by some important recording engineers as the best ribbon ever) is worth taking the trouble to attempt better than 1.7dB NF for its 300R 'nominal' impedance. (the original BBC/STC 4038 were 40R nominal) It is particularly inefficient. IIRC, its 'self noise' is 30dB spl or worse
so every dB of S/N is precious.I wish some of the boutique makers would try making a better, more efficient 4038 instead of that foreign RCA rubbish.
Replacing a 4038's transformer with a zillion ZTX/mA Duraglit would get us at least 2dB if not more.In MicBuilders, I cobble up a dedicated LN ribbon mike preamp from a M-Audio DMP3 ($50-100 on eBay) and my usual string & sealing wax, to get within 1dB of a full Cohen circuit like Millenia Media. NF tells us going the full Cohen monty gets us less than 1dB better noise with a 200R nominal ribbon though Cohan has other advantages
Last edited:
Ok Richard, only to show how confusing your "broadcasting" NF metric is, I'll take the bait again:
1.7dB ref. 300ohm is by your own arithmetic 1.74nV/rtHz (300 ohm has 2.2nV/rtHz), nothing to call home about, actually not something that qualifies as true "low noise". So you are likely including here an unspecified transformer gain. See what I mean?
20*LOG(2.12/x)=1.7dB, renders x=1.74nV/rtHz
I don't suggest you to switch to the En/In noise metric (like everybody in this century) but just to illustrate how the numbers you are throwing around are confusing, due to unspecified side conditions (reference resistance, transformer gains, etc...).
Richard, I don't have models for any MC SUT's but I did run some simulations with some Edcore interstage transformers and found that they ran about 0.5dB "NF" when unloaded driving an ultra low noise FET amp. Did you look at that Jensen app note, ironically it uses my AD743 "DA LOWEST NOISE IC FET OP-AMP IN DA UNIVERSE". I can only figure that they included the op-amp and termination in the noise. Big smilies all around .
.