advertisement


CDX reverse engineering

Teddy - PD is right-on here on all points and yes, your circuit looks like the ones I've found in the CD2 and CD3.5.

I'll try to post my schematics and sims results for the 3.5 later; it shows a very low, constant group delay (bessel roll-off)
 
This is very interesting info. I have a long held hunch that much of the signature sound of Naim players is down to the design of filter they use. I would guess that even small changes here could very noticeably change the end result - for better or worse is hard to predict.

One thing must be acknowledged, Naim certainly are shit hot when it comes to active filters. How many preamps are there that use active filters?

Mr Tibbs
 
Mr Tibbs said:
This is very interesting info. I have a long held hunch that much of the signature sound of Naim players is down to the design of filter they use. I would guess that even small changes here could very noticeably change the end result - for better or worse is hard to predict.

One thing must be acknowledged, Naim certainly are shit hot when it comes to active filters. How many preamps are there that use active filters?

Mr Tibbs
Click to expand...

Lots of older ones?

Remember tone controls - they are just adjustable active filters ;)
 
I guess thats the thing - many of us have found that the bigger gains comes from PSU optimising and not from signal path changes (where a difference occurs not exactly a gain). Good PSUs often bring more of what you liked about the kit in the first place.

Still, three op amps after a modern DAC, i thought newer dacs didnt need as much filtering? Still Naim know theyre stuff i guess.
 
These filter circuits are more or less standard. If you look at the 1702 or 1704 DAC datasheets you'll find similar circuits.

Yes, I agree that PSU makes the largest difference, but the in the case of the CDX, replacing the OP42 by OPA627 was a big difference (improvement...).
 
One thing must be acknowledged, Naim certainly are shit hot when it comes to active filters.
Click to expand...
Tick, agree strongly. Will dig out 3.5 info later.

Trance - on paper maybe, but in fact bistream DACs are noisy, noisy buggers and require a lot of filtering to do the job right. Believe it or not, the amount of energy in the noise from a bitstream dac equates or exceeds the amount of energy in the wanted signal as a result of the way they work! On paper, high OS rates and noise-shaping means that little low-pass filtering is reqd, but in effect - esp. when combined with wide-open amplification - it's just as critical as ever, if not more so. Over many years of meddling I've found that fiddling with things that shouldn't matter (such as such as what the filter does out at 8Fs in the old Marantz CD63 ) can have audible effects when (on first inspection) it really ought not. The key being, no active device is perfect and consequently the best way to bring home the musical goods is in keeping the window open only just wide enough. IMO, of course...
 
Oh and Teddy, I'd agree, and there are reasons for that to do with settling time and bandwidth (not necesarily the same thing..)
 
PD said:
Remember tone controls - they are just adjustable active filters
Click to expand...

Yeah, but I'm talking about bandwidth limiting active filters - only a bat would want them as tone controls ;)


Teddy said:
These filter circuits are more or less standard. If you look at the 1702 or 1704 DAC datasheets you'll find similar circuits.
Click to expand...

In the realm of active filters, similar looking circuits can be anything but similar at handling transients, for example.

A couple of percent change in component values can be enough to change a filter into a completely different breed of filter. We already know Naim like to measure and match components - they may be doing that in critical places of active filters.

Mr Tibbs
 
We already know Naim like to measure and match components - they may be doing that in critical places of active filters.
Click to expand...
Tee hee, i've done the Monte Carlo thing and found (even better) the values Naim pick give very consistent results regardless of typical tolerances for the parts they use. They definitely know what they are doing.
 
A very basic Simetrix model of CD3.5 analogue stage - a rather quick'n'dirty diagram to avoid crossing lines, and using default TL072 opamp(!). The +12v input left represent sthe midrail bias applied to the first opamp (this is a single-rail design, remember) and the AC inut represents the DAC output voltage (TDA1305 is a voltage -output bitstream device):

35_OPS.gif


Frequency response (in green). This displays all ten results of Monte Carlo analysis spanning 10 runs over -/+5% for both resistor and capacitor tolerances. Green line is spread of Group Delay results for same analyses; impressive, no?

FR.gif


Zoom-in on variation in group-delay due to +/-5% in component tolerances: tiny, and well beyond the audible band:

GD.gif


(errors in circuit transcription all mine of course...)
 
martin clark said:
Tee hee, i've done the Monte Carlo thing and found (even better) the values Naim pick give very consistent results regardless of typical tolerances for the parts they use. They definitely know what they are doing.
Click to expand...

I think that is actually a general property of Bessel and other similar alignments.

They have very low Q values, which minimises component sensitivity in the standard Sallen-Key realisation which Naim use.

A filter with a flatter passband (e.g. 6 th order Butterworth) has some sections with Q values over 1, giving rather greater component sensitivity.

I suspect they choose to use Bessel for transient response reasons and get component insensitivity as a nice side-effect, but I may be putting the engineering cart before the horse.
 
Martin,

From your diagram I see that the filter circuit is the same as the on the CDX, there are some differences however:

1. The CDX is split rail, and thus does not require a voltage divider on the op-amp "+" input (but requires a negative regulator...).

2. On the CD3.5 there is an additional op-amp before each filter (X2 and X4 on your diagram), this op-amp doesn't exist on the CDX. Any idea what is the purpose of this op-amp?

3. The DAC on the CDX is 1702, which requires an I/V op-amp, so three op-amps in total.

BTW replacing the first op-amp (the I/V) by op627 gave a very significant improvement, even more than the post filter op-amps.

Which type of capacitors are used on the CD3.5, do they use SMT (ceramic?) as well? Is the 17n2 capacitor in your diagram made of two capacitors in parallel as in the CDX?


Can anyone explain (or send a link) how these filters work?

Teddy
 
Teddy - item 2 models the opamp output within the TDA1305 dac. It has its own f/back cap, for which Naim use a smaller value than given in the DAc's datasheet. It is key to getting the required response.

Using a 627 for I/V will give very good results... I/v conversion exposes the opamp to high slew rates and lots of RF, and requires fast settling.

No Naim didn't invent filtering, the study of which goes back to the first transatlantic cable and Kelvin's neat solution for dealing with the massive capacitance of the line. That doesn't mean that doing filters well isn't an art...
 
I think that is actually a general property of Bessel and other similar alignments.

They have very low Q values, which minimises component sensitivity in the standard Sallen-Key realisation which Naim use.
Click to expand...
Good point. Interesting thing, though, is that the Sallen-key design does rather rely on wide bandwidth in the amplifying devices (IIRC) - maybe one reason better opamps can reap rewards here?
 
For the benefit of other DIYers, the Sallen Key circuit is way of making active filters that uses a single non-inverting amplifier. You need one amplifier per third order filter. Higher Q values require gain from the amplifying stage, which has the effect of making all the tolerances worse. It also requires a flat frequency response from the gain stage - as a rule of thumb the gain stage bandwidth needs to be at least 10Q^2 times the corner frequency for a reasonably accurate response curve.

There is an alternative circuit called the state variable filter. This needs three op-amps per stage (which gives a second order filter), but they only need a bandwidth of about 3 times the corner frequency.

I think op-amp performance in the CD output filters is important because the devices see wideband digitisation noise, which can easily drive the device non-linear. This is particularly true for the first device, the integrator. I suspect the wide band performance of the final op-amp won't be so critical, as the high slew rate junk has already been significantly reduced by the previous stages.
 
teddy_pardo said:
The 2n2 and 1n capacitors are Polystyrene, the others are SMT (Ceramic?), The output capacitor is the usual blue tantalum. All 1K Resistors are 1%, the others are 5%.

My plans are to replace all the resistors by Welwyn, and the SMT capacitors by Polystyrene, any thought on that?
Click to expand...

A message to all CDX owners:

Get rid of these SMT capacitors and replace them with polystyrenes. The difference is very audible!!!

The interesting thing is that Naim thought about it, and there are holes in the PCB to allow polystyrene capacitors. These SMTs probably save some on the cost, but at a cost...

I have also replaced the twelve 1K and the two 5.6K resistors with Welwyn, which is another worthwhile thing to do, but not as much as the capacitors.

Teddy
 
That's interesting - every other bit of naim kit I've looked at uses polystyrene types exclusively for filtering. Yes, it will be worthwhile... and it's been beneficial in all the CD players I've worked on over the years.
 
You must log in or register to reply here.


advertisement


Back
Top