See for yourself - and be amazed, be amazed, because of the paradox that a 150kOhm resistor does not lead to a higher noise voltage! If I haven't made a mistake in the complex impedance conversion, then the spreadsheet in the appendix is valid, it is best to work with S and not jOmega.
R1 is Rin and represents the effective noise current source, i.e. later, or for the real part of the source series impedance, R1 has magically become noise-free and infinite in value, so it is virtually no longer present.
However, this model is only valid if C1 = Cin is practically negligible and the cross-branch of the network can be regarded as purely real, i.e. C in femto Farad.
Good luck!
HBt.
Psst
The 1977 method, with the transformation of the lossy coil into Rp||Lp is the better approach for an intuitive understanding - because now the u^2(tm) = (4*T*k*BW) * Re'p
!
Ultimately, the crucial point is the resulting transfer function of our complex voltage divider, formed from Rs, Ls, Rp and Cp. From left to right and for the noise from right to left or transformed, i.e. by means of noise source shifting ...
Well, are you still following me?
Well, are you still following me?
Below a few nV/rtHz (and no current noise) the contribution from the opamp is dwarfed by the self-noise of a standard 47k load resistor at higher frequencies (where the cartridge L isolates the shunting cartridge R).
Only when that is replaced with a lower noise active version an improvement is seen. But it is brickwalling again when the cartridge R finally becomes the dominant contributor at low frequencies.
OPA1644 sections all in parallel in a 10x pre-pre is quite effective, notably when the ground points of the gain-networks are used as the input current of an inverter for the active load, driving the load resistor.
I suggest the following center frequencies as samples or support points. Why? Because we have to divide the complete audio frequency band in such a way that the later geometric sum of the band noise actually represents df, i.e. 20kHz-20Hz = 19k98Hz as resolution bandwidth.
This is one of the possible solutions.
But please do not forget: Approximations are only approximations and should not replace a correct simulation or a final measurement. They only provide an indication.
tbc
This is one of the possible solutions.
But please do not forget: Approximations are only approximations and should not replace a correct simulation or a final measurement. They only provide an indication.
tbc
Now the preliminary final *.ods version, with 14 sampling points over the entire audio bandwidth - we're mapping the worst case and are being honest with ourselves as designers. For the advertising brochure, we take the red SNR version and perhaps add just 5dB for the A-weighting - ignoring the influence of the operational amplifier.
😉
😉
These appear only available from The Olde Unobtainium Shoppe. My Jurassic lists only have the obsolete AD745 at 3.2nV/rtHz & 7pA/rtHz
Surely in da 21st century, yus LN gurus will know of more?
NJM2068 datasheet dun show separate env & inv. Anyone have more detail on this ... probably the only BJT OPA to slightly beat NE5534
What's wrong with the IEC 1/3 octave spacings 20 25 31.5 40 50 63 80 100 126 160 200 etc. I think Nick & Marcel sorta use these.
I use them cos B&K 2307 chart recorder toilet roll has special marks for this eg https://www.proaudiodesignforum.com/forum/php/viewtopic.php?t=423&hilit=ricardo+riaa&start=30
I dun use da 2307 anymore cos I can't get the steam in Cooktown but I've laid on a year's supply of B&K toilet roll in case of cyclones or another pandemic 😊
A point of note, the AD8656 (which is only 5V supply) is CMOS, quotes 2.7nV/√Hz (but for 10kHz, its 4nV/√Hz at 1kHz), and 1pA bias current. So the 1/f knee is high, but CMOS is where to look for new low noise devices perhaps, not JFET. For instance the LMR1802G-LB quotes 2.9nV at 1kHz, they are slowing eating away at the 1/f knee for CMOS. Assuming you can live with +/-2.5V rails...
Nothing!
But how are they supposed to be useful as narrowband measurement filters? How can you be sure that they correctly map the noise content at this point (which is their center frequency)? Are they rectangular windows or ideal bandpass filters? How do you weight each of these sections - and why? Does the integral divided by X(end)-X(start) really represent the noise voltage in the entire audio bandwidth?
Last edited:
They are centre frequencies for a simulated 1/3 8ve RTA eg https://www.bksv.com/media/doc/bp1021.pdf
Can you explain how your 14 frequencies do this ... while IEC 1/3 8ve frequencies do not?
In a spreadsheet sim like yours, they are rectangular. If you are analysing a REAL noise measurement, each FFT bin is a Sinc (modified by the original 'windowing') But if your FFT is large (>2048 pts), this doesn't really affect your results ... even at the lowest frequencies when only a few FFT bins are in the "1/3 8ve"
As I said earlier, I'd much rather see the spectrum (as a 1/3 8ve display or finer cos the log frequency scale) after RIAA EQ. But for your "single number" calculation, I'd apply CCIR 468 2kHz before RMS summing the lot.
Thanks for this Mark. I'm rather surprised no one has made a cheapo OP627 with <3nV/rt(Hz) but still with +/-15 supplies. Perhaps the current requirement was a problem; OP627 already taking 7mA for 4nV/rt(Hz)
The late Guru Wurcer told me that, once a process was selected, it was just a matter of allocating enough area for the input FETs. Indeed this is probably how NJM make a whole bunch of BJT OPAs optimised for different applications using the same old topology which I first encountered in Raytheon 4136
BTW, going through Wayne's thread on ProAudioDesignForum and chasing up half remembered links there,, it appears NJM2068 has less env than LM4562 ie < 2.5nV/rt(Hz) at 1kHz and sufficiently low inv to better 5534 for MM RIAA
I'm sad you think improved LN stuff will only appear on +/- 2v5 rails. My history with RIAA preamps starts with attempting to integrate early LN transistors (eg BC109) into hybrid valve circuits.
Then early solid state preamps with 2 or 3 transistors were limited to a single 20V supply which had serious problems with overload margin.
Later 'HV' LN transistors allowed sensible >100mV 1kHz input overload for RIAA with OPAs on +/- 15V juu.uust allowing +20dBu output with careful design.
Overload Margin is important for anything before the main Volume Control or Fader. Operating at -10dBu for domestic stuff and 0dBu for 'professional' stuff allows 30dB and 20dB Overload Margin respectively. Essentially how low you can set your Fader or Volume Control so the earlier stages don't overload before the 'Power Amp'
Da +/- 2v5 stuff is meant for portable devices, often when the 'input' stage can have it's gain set digitally so no Overload problems. Alas it is difficult to do this with accurate LN RIAA so we've loat 15/2.5 = 15.6dB Overload Margin immediately; worst than the early 20V transistor designs
Gotta do a drastic rethink on levels & topology with these
1/3
8ve
are not narrowband-measurementfilter !
[td width="86px"]
20
[/td][td]
25,1984209979
[/td][td]
31,7480210394
[/td][td]
40
[/td][td]
50,3968419958
[/td][td]
63,4960420787
[/td][td]
80
[/td][td]
100,7936839916
[/td][td]
126,9920841575
[/td][td]
160
[/td][td]
201,5873679832
[/td][td]
253,9841683149
[/td][td]
320
[/td][td]
403,1747359664
[/td][td]
507,9683366298
[/td][td]
640
[/td][td]
806,3494719327
[/td][td]
1015,9366732597
[/td][td]
1280
[/td][td]
1612,6989438654
[/td][td]
2031,8733465193
[/td][td]
2560
[/td][td]
3225,3978877309
[/td][td]
4063,7466930386
[/td][td]
5120
[/td][td]
6450,7957754618
[/td][td]
8127,4933860772
[/td][td]
10240
[/td][td]
12901,5915509235
[/td][td]
16254,9867721544
[/td][td]
20480
[/td]16k255 -> 14k482 till 18k246 -> df=bw= 3.764kHz
25.2Hz -> 22.5 till 28.3 -> df=bw= 5.8Hz
RTA -> used pink noise, not white noise ..! Do you know where the journey is going now? Spreadsheet calculations are rough and clearly error-prone pocket calculator methods. Or to put it another way: the geometric sum of the individual noise of the bands (better formulated as the so-called interpolation or sampling points!) must not fall below 19.xx nV as the equivalent total noise voltage in relation to 1 ohm.
Otherwise the pocket calculator method will spit out values that are too good. A word about the rating curves: in the first instance, we can omit all filters in this regard.
"KSTR" Has already addressed the inherent problem of the MM pickup. In a nutshell, the system itself sets a limit - and we will never be able to achieve fantastic SNR figures.
Typo
"not below 18.xx nV"
This snapshot better illustrates the hidden weaknesses of the “pocket calculator method”: df becomes too large at some point to match the selected sampling points and no longer represents a narrow band.
tbc
The alternative and very direct method - hidden in the spreadsheet, the more supporting points, the better. The maximum achievable gain from the RIAA weighting is 12dB, so you can simply add it on top at the end. The icing on the cake would then be 5dB for the A-weighting.
The greater the difference between 150k and 47k, the more skeptical you have to be. I wrote about the possible differences earlier in this thread. For me, the case is now closed and I will return to the beginning, i.e. I will edit the win32.exe so that the two most popular topologies and the pickup are correctly taken into account.
HBt.
The greater the difference between 150k and 47k, the more skeptical you have to be. I wrote about the possible differences earlier in this thread. For me, the case is now closed and I will return to the beginning, i.e. I will edit the win32.exe so that the two most popular topologies and the pickup are correctly taken into account.
HBt.
Last edited:
*.ods
For general disposal and completion, accompanying you on your way forward.
Disclaimer
I may have made 1001 mistakes, this is especially easy with an excel-sheet. I assume no liability for any consequences.
For general disposal and completion, accompanying you on your way forward.
Disclaimer
I may have made 1001 mistakes, this is especially easy with an excel-sheet. I assume no liability for any consequences.
add to my posting #135
To memorize, we can never get better than 81dB, our amplifying element will degrade this given limit.
Case closed!
HBt.
To memorize, we can never get better than 81dB, our amplifying element will degrade this given limit.
Case closed!
HBt.
Thanks KSTR and YES to all that.
I'm not really (well only very slightly) interested in RIAA with more than 1 OPA. The big draw for me is that the above Unobtainium FET OPAs might significantly outperform 5534 or even NJM 2068. Dis beach bum gotta get his finger out and do his own spreadsheet & fancy curves to check on this.
anyone help a lazy beach bum find da best active load RIAA preamps ... hopefully with measurements and details of how they deal wid da cartridge maker's loading recommendations?
There are enough current-production JFET opamps that are significantly better than those bipolars and the OPA627/37 (4.5nV/rtHz):
OPA827 and OPA828, ADA4625, ....
OPA828 is a perfect choice, IMHO. Low noise, high bandwidth/gain, very low offset. It's the official replacement for 627/37, after all.