Accurate Inverse RIAA

www.hifisonix.com
Joined 2003
Paid Member
I've built, tested, and written up an article for an Inverse RIAA Network (IRN) - you can down load the article from my website here

An Accurate Inverse RIAA Network

You can now buy PCB's for this board - just follow the link above. The Gerbers are available at the link above as well for the diehards who want to etch their own.

Do not use the Gerbers to manufacture and sell these boards. they are copyrighted to hifisonix

Having built and used this on some RIAA EQ amps that I am working on, I cannot recommend this simple tool enough for RIAA EQ designers and especially so if you use square wave testing to assess conformity - I discuss this in the article.

Acknowledgements: The design is based on the Reg Williamson IRN from TAA 1971 with the Jung Lipshitz mods also published in TAA in 1979
 

Attachments

  • photo.JPG
    photo.JPG
    193.8 KB · Views: 883
Last edited:

PRR

Member
Joined 2003
Paid Member
Minor typo: "Response Conformity 20 Hz to 20 kHz Worst cast"

"Use of 600 Ω Source impedance can degrade accuracy by a factor of 3 worst case"

Factor of 3...what? 3dB? 9.54dB?

AFAICT: "worst case" would be up toward 200KHz.
 
Agree that a good inverse network is essential for testing, but I'm not grasping how paralleling more parts necessairly improves tolerances. If you need a 200 gallon capacitor and have a 200 1% on the high side, it's 202 gallons. If you use two 100 gallon parts, both high by 1%, you still end up with a 202 gallon part in parallel. I think if you have a statistical distribution things improve, but I've got many bags of components where all are a bit higher or a bit low. Worse, some have had the closest values removed by the manufacturer to fill the next grade up, so the bag has a double hump distribution. IMO, it would be better to just select cap values and add some trimmers to fine tune to non standard resistance values.
 
Agree that a good inverse network is essential for testing, but I'm not grasping how paralleling more parts necessairly improves tolerances. If you need a 200 gallon capacitor and have a 200 1% on the high side, it's 202 gallons. If you use two 100 gallon parts, both high by 1%, you still end up with a 202 gallon part in parallel. I think if you have a statistical distribution things improve, but I've got many bags of components where all are a bit higher or a bit low. Worse, some have had the closest values removed by the manufacturer to fill the next grade up, so the bag has a double hump distribution. IMO, it would be better to just select cap values and add some trimmers to fine tune to non standard resistance values.

That's also my experience, especially with resistors.
I agree.
If one were to measure all the 1% and found a total range of 2% and selected from either side of the median, then you will get a more accurate combination.
I have never seen a 2% range for 1% components (nor a 10% range for 5% components), so we can't reliably take advantage of the statistical averaging.

Just measure as best as you can and if possible compare to some accurate low tolerance specials. I bought a few values of ±0.3% capacitors so that I could "roughly" calibrate the nF scale of my DMM. But I would not claim better than 1% absolute.

Buying from different manufacturers should not make any difference. If you are buying ±1% tolerance values, then all in specification devices are inside the ±1% tolerance range.
 
Last edited:
www.hifisonix.com
Joined 2003
Paid Member
1. This technique assumes a normal Gaussian distribution - quite an ok assumption with modern electronic components manufactured in high volume.
2. Component manufacturers invest a huge amount of effort in getting good repeatability and tight spreads. Why? Because they use this information amongst other to drive cost reductions
3. Specifically for the paralleling or series technique, the following holds true, and this is why we use it:-

"To make up a desired value, if you parallel or series components of the same value and that have a normal distribution, you will almost certainly get a tighter tolerance in the overall combination value than the specified worst case tolerance of the individual devices"

As with a few others on this forum, I worked in semiconductors for 20 years. All hell broke loose when parameters drifted off their nominally specified values. In high volume production, you have to take these things seriously. Try shipping 1 million parts to an auto manufacturer like Bosch with a key parameter that's sitting on the edge of its specified tolerance band.

Separately, make sure you buy parts from reputable suppliers. Stuff off of traders on ebay and Amazon is probably not sourced from the right places, and may in some cases be production rejects.
 
Hi,

I build myself this network in 2009 and used it often since.
Its stereo, switchable between MM (-40dB/600Ohm) and MC (-60dB/60Ohm) and allows for four generator impedances (0R 50R 550R 600R).
You can feed it from a CD-player and connect it to Your Phono-preamp to directly compare the original (CD-)sound vers. Your Phono-pre sound.
Can be a quite disillusioning test for many phono-stages. :cool:

jauuu
Calvin
 

Attachments

  • inv RIAA-Neumann sch.jpg
    inv RIAA-Neumann sch.jpg
    149.6 KB · Views: 1,036
  • inv RIAA-Neumann brd.jpg
    inv RIAA-Neumann brd.jpg
    81.6 KB · Views: 995
  • inv RIAA-Neumann PCB.jpg
    inv RIAA-Neumann PCB.jpg
    49.3 KB · Views: 913
  • inverse RIAA - amplitude MC.jpg
    inverse RIAA - amplitude MC.jpg
    153.1 KB · Views: 899
  • inverse RIAA - amplitude MM.jpg
    inverse RIAA - amplitude MM.jpg
    150 KB · Views: 310
  • invRIAA 01 small.jpg
    invRIAA 01 small.jpg
    70.8 KB · Views: 385
  • inv RIAA-Neumann print.jpg
    inv RIAA-Neumann print.jpg
    80.8 KB · Views: 444
1. This technique assumes a normal Gaussian distribution - quite an ok assumption with modern electronic components manufactured in high volume.
.

It also works with a uniform distribution, though you're not likely to encounter one. See Small Signal Audio Design 2nd edition, Chap 2.

I have never yet come across component batches where the centre 1% (say) has been selected out. I can't see that this could be economical for something like resistors; if you want 1% you make 1%.
 
www.hifisonix.com
Joined 2003
Paid Member
It also works with a uniform distribution, though you're not likely to encounter one. See Small Signal Audio Design 2nd edition, Chap 2.

I have never yet come across component batches where the centre 1% (say) has been selected out. I can't see that this could be economical for something like resistors; if you want 1% you make 1%.

Thanks Doug - yes, I was thinking about uniform distribution today and came to the same conclusion. I you happen to pick a component that is at the very limit of its specified tolerance, you'd have to have extraordinary bad luck to pick a second with the same tolerance. Much more likely you'd pick a closer tolerance part and the overall tolerance of the network would improve when you combined them.

Of course, the reverse also works, but on average, paralleling (or placing in series) tightens the tolerance of the network.

Thanks BTW for bringing this technique to the fore in SSAD.
 
www.hifisonix.com
Joined 2003
Paid Member
Just read your website Bonsai. Wonderful place! Do you happen to sell boards for your designs or does anyone else?

Hi - thanks for the feedback.

I don't provide boards, but you can buy very high quality DSTHP boards for the sx-Amp and the nx-Amp from Jims Audio. If you look at the first post in the sx and nx Amplifier thread, all the details are there.

http://www.diyaudio.com/forums/solid-state/236522-sx-amp-nx-amp.html

There are no boards available for any of the other projects.
 
Member
Joined 2005
Paid Member
Hi,

I build myself this network in 2009 and used it often since.
Its stereo, switchable between MM (-40dB/600Ohm) and MC (-60dB/60Ohm) and allows for four generator impedances (0R 50R 550R 600R).
You can feed it from a CD-player and connect it to Your Phono-preamp to directly compare the original (CD-)sound vers. Your Phono-pre sound.
Can be a quite disillusioning test for many phono-stages. :cool:

jauuu
Calvin

beautiful and well thought through, IMHO. agree on the disillusionement, been there....
 
www.hifisonix.com
Joined 2003
Paid Member
What about modeling the inverse RIAA in LT Spice (voltage controlled voltage source with Laplace transform to create the inverse RIAA), input a WAV (white noise or whatever you need) and "record" the WAV at the output, and use these resulting files for testing?
Some hints:
http://www.diyaudio.com/forums/analog-line-level/74983-spice-transfer-function-riaa-testing-2.html

Using WAVE files as input/output in LTSpice | ElectroStud | Fandom powered by Wikia

That should work very well.
 
Hi,

You'd have a precise inverse RIAA wav-file, but at what signal- and impedance level?
Feeding from a typical Preamplifier, it'd be almost impossible to set the volume level right and consistantly right to not to overdrive the Phono input.
You'd need at least a voltage divider between preamp out and phono in with considerable attenuation, and where the resistance to gnd is of roughly the pickups value.
The hardware inverse RIAA still has the advantage that it can be used with any input signal, be it a test signal from a generator or music from a highlevel source.

jauu
Calvin
 
Ok, so such inverse RIAA network feeding a phono stage effectively comprises a flat line level stage.
Of course such attenuation and then phono gain will give worse SNR than a normal line stage.
This would provide a ready way to discriminate the noise characteristics of different phono stages.
What are your guys findings in listening to different phono stages by this method ?.

Dan.
 
www.hifisonix.com
Joined 2003
Paid Member
Hi,

You'd have a precise inverse RIAA wav-file, but at what signal- and impedance level?
Feeding from a typical Preamplifier, it'd be almost impossible to set the volume level right and consistantly right to not to overdrive the Phono input.
You'd need at least a voltage divider between preamp out and phono in with considerable attenuation, and where the resistance to gnd is of roughly the pickups value.
The hardware inverse RIAA still has the advantage that it can be used with any input signal, be it a test signal from a generator or music from a highlevel source.

jauu
Calvin

Yes, I agree - I think the hardware IRN is very versatile.

I believe Scott Wurcer has done some work on software based RIAA EQ - you could do exactly the same for IRN.

I like the hardware IRN because you can pass a CD signal through it and then get a normal signal out of the RIAA EQ that you can directly compare to the original. This is a tough test as noted by someone a little earlier. See Denis Colins LP797 RIAA where he does this (he calls it an 'intrinsic fidelity' test).
 
I like the hardware IRN because you can pass a CD signal through it and then get a normal signal out of the RIAA EQ that you can directly compare to the original. This is a tough test as noted by someone a little earlier. See Denis Colins LP797 RIAA where he does this (he calls it an 'intrinsic fidelity' test).
Yes, that is my question as per my above post.
What do you find subjectively ?.

Dan.