Meridian Unplugged - Powered by MrTechGuy.co.uk Forums Discussion Meridian - Systems All FIFO buffers are not the same. Discuss...

Hi Mark,

I'm trying to make sense of Meridian's claim that their new SpeakerLink is better than coax. The *only* reason I can think of is that the FIFO buffering in DSP speakers is not 100% effective. Would you agree?

If so, I'm wondering what it is about the speaker FIFO that perhaps differs from FIFO in the processors. (As an aside, based on some of the "bits are bits" arguments we've had on this forum, I have suspected for some time that the FIFO in DSP speakers is not as effective as that in the latest processors. As another aside, I noticed that the FIFO code is versioned; G68 FIFO is currently v1.4.)

I wonder if DSP speakers simply *cannot* afford to lose FIFO lock as they don't have a backup (processors have the XMAS mode - a pseudonym for using the source clock IIRC). I'm not sure about this, but if so, could this mean that the number of levels of buffering and/or the depth of the buffering at each stage and/or the overall jitter transfer function is different for the speakers' FIFO?

Hi Joel,

Sorry for the essay which is about to follow!

About the 100% effective thing, yes I'd have to agree.

You're right on the limitations on the speaker FIFO being different, too. Your point about the fact that they can't afford to lose lock is definitely pertinent, but so is another (that hadn't occured to me before). They can't have large buffer depths. Whatever distance the buffer is away from 50% [buffer fullness] is delay, either early or late, but an offset with regards to other speakers (unless they all have precisely the same depth at that instant). Since localisation is all at the high frequencies, and a delay by a couple of samples is enough to competely screw up the phase of anything over 5kHz or so, then if all speakers had different depths, the imaging would be terrible. Then again, this is nothing compared to the latency range that DSPs used to have! A high jitter source should test this more than a low jitter one.

This would be cancelled if the current buffer depths were identical. This is only possible if they're fed from the same clock source (which I guess they are), and their PLLs all have the same behaviour. That's hard to guarantee. Especially across devices (if they go obsolete).

Your point about losing lock requires a PLL with a relatively high bandwidth. So does the other point about limiting excursion in the buffer. A high bandwidth PLL is like a flywheel with less mass. It responds quicker (enhancing lock range), but filters out less noise (jitter).

In summary, I 100% agree with you.

Oh here's another thought! And it's a good one. I'd assumed that the reason Cat5 was better was because it's balanced (which is true). What if the processor generated a master clock itself which was then distributed to the speakers? It would be low jitter because it would be a pure square wave, not manchester encoded like SPDIF (which introduces seriously bad data dependent jitter). With the better quality clock, all speakers should have less work do do in the jitter filtering. That should mean that a lower bandwidth PLL can be used whilst still controlling excursion and locking. This means a better quality clock at the DACs, and less time difference between speakers.

Fantastic. Finally, this is an explanation as to why really low jitter (exotic / expensive / Meridian) transports sound better *direct* to Meridian FIFO DSP speakers than higher jitter transports.

In other words, those who claim the 808.1 [90ps jitter] sounds better than a Squeezebox into their DSP8000s are probably right. However, those comparing an 808.1 vs. SB into a G68 and *then* DSP speakers will probably *not* hear a difference (at least not in a double-blind test).

There is a further point... An 808.2 -- SpeakerLink --> 7200s *might* sound better than 808.2 -- SPDIF --> G68 -- SPDIF --> 7200s. It all depends on the output SPDIF jitter from the G68.

Agreed again - it certianly would explain those differences.

It's funny that discussing SpeakerLink has possibly given us a big insight into an old chestnut. For me, the last bits-are-bits argument never reached a satisfactory conclusion. I must confess that, whilst I (still) believe (strongly) that most people do not know how to audition/compare equipment and are strongly swayed by psychological factors that they are not even aware of, there were some opinions and reports of tests that I just couldn't dismiss. I need to go back and re-read those threads, but it seems to me quite likely that we missed the critical point that "all FIFOs are not the same".

(We do still think that the processor FIFO is more effective, right?)

Unfortunately, there are so many possible digital paths from source to speakers (upsampling, res enhancement, FIFO, etc.) that apples vs. apples comparisons are really difficult in a public discussion, particularly when the definition of apple hasn't been nailed down!

I'd guess that the processor FIFO is different because of the different constraints. The main limit on the change in depth will probably be lip syncing issues I suppose. Here's another thought, though. Since the bandwidth is effectively controlled by firmware, the Xmas option could actually enable a higher bandwidth, rather than directly recovering the clock. If that is the case, Xmas is likely to be a high bandwidth mode for the FIFO. This could be checked by sticking the diags screen on the G68, configuring an 'average' jitter source (Oppo?) to two sources, one of which is 'Poor Digital'. The FIFO screen should still be there.

Yes, I guess it would be much much simpler from an engineering perspective to simply change the bandwidth in the FIFO than re-route to a completely different de-jittering or pass-through mechanism.

All FIFO Buffers are NOT the same

In its own 861 data sheet, Meridian claims that the 861 has "a memory-based de-jittering system that renders the processor almost entirely immune to input-signal quality". Presumably this applies to the G68 too.

IIRC, the G98 doesn't upsample and I don't think it outputs RBCD using MHR.

Although producing sources with low jitter is a good thing and makes for the best compatibility with the widest range of partnering equipment, jitter does not affect digital processing and can, in fact, be minimized/removed at ANY stage prior to the DAC