Is the SAA7220P/B really that bad ?

Hi,

Well I'm hoping that the title of this thread will stir up some debate !

I am expecting a lot of 'yes it is, it's a bad chip, rip it out and go NOS' but I'd like to understand the reasons behind people thinking NOS is better.

There will be some, I'm sure, who think that interpolation is just wrong and with any oversampling filter you're not listening to the real music. There will be others, who have tried it and prefer the results because it sounds better to them; whether the sound is more 'realistic' or not.

What I am wondering is, why this is? Is it, not because the digital filter is a bad chip, but, actually because it creates a lot of pollution on the power rails which affects the DAC; with the data clocked at 4fs instead of 1fs timing is more critical (jitter) and since everything is running faster more attention needs to be paid to design of the power supply, decoupling, layout, grounding and so on. If all these things are taken care of can the combination of SAA7220P/B + TDA1541A work well ? Is it really that the implementation of the SAA7220 in most players leaves something to be desired ?

Cards on the table. I have an Arcam Alpha 5 (SAA7220P/B + TDA1541A) and I have socketed both chips so I could easily go NOS and I will and compare the result but if I do like it better I'd like to know why, certainly reproducing a 20KHz sinewave with a 44.1KHz sample rate isn't going to sound good !

So, let the comments begin ...

Regards,

Jon
 
I just yanked out SAA7220 and replaced it with Net-Audio NOS board .Well, it did improve the sound considerably (there were other changes in the player as well ) but judging from the effect of the change it just makes not much sense to me doing clock upgrade if you keep SAA7220 in place. I don't care for NOS or no NOS but clean and properly implemented clock gotta make a difference and what I gather there is nothing you can do to improve on SAA7220. The board isn't cheap at 40 GBP but if I was to make only single mod on my player it would be replacing the filter chip with it. Regards, L
 
Good idea for a thread!
Is it, not because the digital filter is a bad chip, but, actually because it creates a lot of pollution on the power rails which affects the DAC; with the data clocked at 4fs instead of 1fs timing is more critical (jitter) and since everything is running faster more attention needs to be paid to design of the power supply, decoupling, layout, grounding and so on.
You've nailed it. The SAA7220 is a noisy beast, and draws quite high currents (c180-200mA). The problem with that is simply the size of the package - you have the digital filter and the clock generator and clock divider all sharing the same power supply pins, which are so far apart that true HF decoupling is difficult to do well (opposite corners of a large DIL package). That gurantees a mess on the jitter front. So there's a two-part fix:

1) Clean-up the PSU, develop decoupling strategies to minimise noise on the supply pin.
2) Simply stop putting the clock signal through the filter. Use a separate 11.288Mhz clock with D-flipflop divider* to feed 5.66Mhz to the 1541, and 11.288Mhz to the 7220 clock input and to the preceeding 7310 decoder. That is - take all responsibility for clock generation and distribution away from the 7220. Problem solved!

If all these things are taken care of can the combination of SAA7220P/B + TDA1541A work well ? Is it really that the implementation of the SAA7220 in most players leaves something to be desired ?
Definitely yes to both parts, for the reasons above. Now - is it the best you can do with a 1541? maybe not, but certainly I don't feel I'm missing anything from my current player with this setup.

Do try both methods with you Arcam 5 and see what works for you. I think that the reason NOS remains popular is because it's much easier than 'sorting' the SAA7220 (which is incredibly context sensitive and requires a scope etc). Having tried it both ways, I know that the NOS sound definitely isn't for me.
 
One more little thing - for chips designed to work together - they don't interface very well without a bit of work! Find the long 'TDA1541 thread in this forum of more details, but the key tweak is to use RC decoupling in the signal lines between these two chips to reduce HF noise injection into the dacs substrate (1K/10pF was suggested)*. Your Arcam already has some resistors in place in these lines, so definitely add small caps (10-22pF) to digital ground at the 1541 input pins. It helps a surprising amount.

*this is because the 7220 swings 5v pk-pk on its outputs, and the 1541 only needs a tiny current centred c1.4v to switch - it's a current-routing logic; the excess voltage swing just pushes HF currents into the dac.
 
Martin,

I like the idea of just using the SAA7220 as a filter and removing the other functions. A seperate clock 'factory' with buffered feeds to the A and B chips and the DAC sounds like a very good idea.

RC in the digital lines is certainly a good idea. You could certainly get away with slowing those edges down a bit, looking at the datasheets I don't think you'd want to go over 22pF though. Reclocking the signal lines should help as well but don't choose logic that is too fast !

Ashley,

I don't know what the attenuation is. The datasheet doesn't go into details about the type of filter; beyond stating that it is a 120 tap FIR no other details are given. I'd guess that with that number of taps it might be anywhere from say 20 to 80dB depending upon the type of filter.

All,

The grounding scheme on the Alpha 5 output board is really nothing to write home about. Doing another board layout would be the best option but I'm not doing that. I did think about fitting a sheet of self adhesive copper underneath the pcb (adhesive backing still in place) and connecting down to that.


Keep the feedback coming !

Jon
 
Hi Jon and Martin

Have a look at the audioboard in the Alpha 5+. The clock looks a bit better than in the Alpha 5. It is generated outside the SAA7220 and all lines going to the DAC are reclocked. The only downside is that all of the reclocking is done in one quad filp-folp.

A similar approach was taken in the Delta 70.2 but only the bitclock was reclocked in a dual flip-flop (along with the signal for the digital output - which can easily be disconnected to optimise the reclocking of the bitclock).

The ideal situation is obviously to have a clock distribution with 5 outputs (one at half fequency). You could then directly clock the dac with the 1/2 clock, reclock WS and DATA using individual flip-flops and separately clock the SAA7220 filter and SAA7210/7310 decoder.

The closest ready made solution I have found is a Tentlabs XO 2.5. This provides three normal outputs and one 1/2 frequency output. The 1/2 frequency output would no doubt be used to directly clock the dac chip. This then poses the question of which is most improtant to clock/reclock with the remaining three clean clock outputs. Is it better to:

A. Reclock WS and DATA using separate filp-flops and feed the final output to the SAA7220......and let it then feed the decoder.

B. Separately clock the SAA7220 and decoder.....and use the final output to reclock WS or DATA (or both in a a shared package).

C. Separately clock the SAA7220 and decoder.....and not use the final output.
 
Hi All,

A good thread indeed.
My only experience is with NOS (I chose it because, at such low frequencies, it should be easier to do as a first DAC), and I want to try oversampling because I want to be able to compare.

For me, I think the biggest problem is finding an SAA7220. Are there any obvious, old donors out there, expecially with a DIL version of the chip?

But, I did always wonder, is the 7220 frowned upon because of its typical electrical environment, or is it because there are perhaps better digital filters out there? For instance, how would a 7220 fare against an AD1896?

Cheers,
Phil
 
Why is it an invalid comparison?

I know that the AD1896 is an ASRC, and can do a lot more than just oversampling, but surely it could output 16 bit 88.2 or 176.4 kHz I2S?
Isn't that equivalent to 2x or 4x oversampling?

Can't that be compared to an SAA7220? Just like if you got a TMS320CXX to do the same with a specific DSP algorithm and compared?

Or do you refer to the supporting electronics required with more advanced chips that can seriously impact performance due to high bandwidth noise pollution on the power rails, the need for more careful PCB design etc?

I've never done oversampling before, and so that's why I ask.
 
philpoole said:
Why is it an invalid comparison?

I know that the AD1896 is an ASRC, and can do a lot more than just oversampling, but surely it could output 16 bit 88.2 or 176.4 kHz I2S?
Isn't that equivalent to 2x or 4x oversampling?


In theory I can tow a plough attachment with a Freelander but that doesn't make it a tractor. The AD1896 can raise the sample rate but that does not make it the same as the SAA7220. They are aimed at different tasks.
 
Interesting thread. I did something along these lines in my Rotel... clocked the 1541A DAC directly off of a modified version of one of Martin's Flea boards with a /2 added. I haven't reclocked any of the other lines. The Tent XO module clocks the 7220, which distributes clocking to the 7210. Works well, though I did have to toss an inverter in the /2 circuit to get the timing to work. I would be interested in hearing about reclocking the other signals.
 
Hi All,

Thanks for the links. Looks like I might be placing an order with partsmaster. I'll let you know if they're any good.

So the point with the AD1896 is that it is simply overkill. Fair enough, but it could be configurable to do the same as an SAA7220, and in that mode I was wondering if either of them behave as well.
I agree it would be better to use it to upsample to something like 192kHz at 24bit, but I was just trying to understand if the two can perform oversampling in similar ways, with similar results.

I think this has established for me that the 7220 is a perfectly capable chip, but really needs a good implementation.

Cheers,
Phil
 
philpoole said:

So the point with the AD1896 is that it is simply overkill. Fair enough, but it could be configurable to do the same as an SAA7220, and in that mode I was wondering if either of them behave as well.

The point is the AD1896 makes for a crap oversampling filter. The SAA7220 is a 4x oversampling digital filter. Its coefficients are chosen and optimised for that oversampling ratio. Moreover it is a synchronous device.
 
I tried nonos with removing a SAA7220 in a CD640 with TDA1541. At first it sounded good but with longer listening i got tired of it.

With a CD304mk2 with nice solid CDM1 dicided to try to get its best of a 7220, left that by some here called "stinker" at its place, but did a lot of other (standard) mods:
Better decoupling with a 3rd order PS filter to 7210/20 (Sanyo oscon, bead and tantalium)
All caps upgraded to newer & better ones. In dig. PS i added a inductive coil between the two lytics after rectifier.
1N4004's out, BYV28 in with 1 ohm resistor in AC line to stop further noise.

Also the stopping R's between SAA7210,7220 and 1541 are not forgotten in I2S lines.

Best would be a separate heavy shunt supply to the 7210/20. Rudo did a great job with a separate shunt ps for 7220, see:
http://home.hccnet.nl/r.r.meijer/cdd.html

I want to try it, never done so far.

But what i further did to improve the CD304:

Clock: Kwak 7 with 11Mhz separate to SAA7210 and SAA7220, and /2 divided clock (with 74/163) to the tda.

Opamps upgraded to THS4032, on a Dil socket,soldered with 100N mks cap for filtering against oscillations. PS to opamps also a sort of 3rd order filter with standard philips lytic, bead + 1ohm, and close to opamp a 1uF MKT cap. The cdm uses same +- 12V, now i cannot hear the servo seeking anymore between tracks!

Output caps is done with a motor starter cap, an oil dampened 6.8 uF MKP for cheap.

Now i have a fine sounding player, and can finally judge now the 7220 is not the greatest stinker anymore. The 304 sounds allmost like nonos (natural) but i am not getting tired anymore of the sound as with nonos, can listen for hours and hours.