Meridian Unplugged - Powered by MrTechGuy.co.uk Forums Discussion Meridian - Systems M really can't do video!!!

Are these in-house algorithms where the chroma bug comes from?

Mark S

No, the chroma upsampling error (CUE) occurs inside the MPEG decoding chip. For some reason (possibly early DVD-A support), Meridian chose the C-Cube Ziva 3 (596 and 800) and Ziva 5 (598, G9x and presumably the 800v4). Luckily, the CUE bug can be masked, but it's still not ideal.

Meridian is not using fully programmable processors. The only thing that is "programmable" are options to turn certain features on or off and making some adjustments in processing.

I asume you're talking about FPGAs when you say fully programmable. If that's what you're talking about then you have to look at the big Teranex. If you open one of those up and pull some boards, you'll see why fully programmable is not an option for Meridian yet. It requires a huge amount of processing power, since FPGAs are not very effective in comparison to ASICs.

What Meridian is using right now, are standard processors that were built by Faroudja. Just get your hands on one and check what's inside. And while you're at it, put it side by side to the usual suspects (Lumagen, DVDO, ...) and see how bad the Faroudja/Meridian performs. Don't forget to compare the 48Hz output...

Seriously, Meridian is far behind when it comes to video. I've talked to people at Meridian (engineering and sales) and asked about certain things ever since the first prototype VE12 board appeared. Back then, even engineering didn't know about certain features the hardware was capable of (it wasn't documented by the chip manufacturer for whatever reason). I've tried to point them in the right direction several times, but with no success.

I'll talk to Ron again in May and see what their plans are for the near future. But I honestly don't have high hopes for anything new.

If they're working on a new processor, that's good for them. I just hope they will implement all features that people want. With their currently available processors, they're stuck in a dead end.

Well I guess M UK is no different than M usa- Des just sent me an e-mail saying that they will get back to me in due course???Not sure what the heck that means?? I must say that all this runaround and no communication is very disturbing- I wish they would just give me my money back -I will take my losses and move on- gary

Stephan

I suggest you tell this to Bob Stuart then

I think he knows what they are using and basically he stated to everyone at the open day that it didn't matter that other companies were touting that they has this chip or that chip since their video solutions were not dictated by the capabilities of a standard chip and they were fully programmable and updatable to the very latest algorithms that Meridian develop in house.

I think this is the case across the board from G series and beyond. At last years Bristol show, Bob Stuart commented that the deinterlacer of G9x and Meridians top of the range would produce identical results due to the programmability. Where the G9x could not compete was with scaling as the dedicated boxes had more DSP power available to it do could do a better job.

I'm going off topic, but I'm going to defend the move to FPGA. I think Stephan raises some valid points, but I disagree on the FPGA move. I'll also mention up front that I'm not impartial - I design FPGAs for a living.

You're going back a while there, Stephan. ASICs nowadays are just for high volume consumer stuff (even then only for when they can be amortised across multiple products). You can run FPGAs at 600MHz and have millions of gates available. MRI scanners and 4D ultrasounds use them for real time processing - if there's enough power for that, there's plenty for video processing. All the big telecomms boys have been developing mostly on FPGA for years for all their really powerful stuff, reserving ASICs for handsets where the functionality is fixed and volumes are in the millions. It's just not practical to develop high performance ASICs at 65nm unless you know you have a huge market. Mask costs are silly. It takes a year to design (at least), then another year to verify - how do you stay cutting edge with that? That's why I thing the Genesis tie up is so good for them - they get functionally verified code.

Application specific stuff like Genesis is great when you can share identical functionality across lots of devices, but you can't differentiate with a chip that everyone else has got. If Meridian are using FPGA, then that is a great strategic move. We're all used to software updates - now we'll get updated hardware as well.

The bottom line is that all but very few can afford to do latest process ASICs. In a gate for gate per-die comparison, FPGAs cost a fortune. In return you get real hardware future proofing and ludicrously high performance versus DSPs. At best ASICs are 50% faster - that can easily be overcome by paralleling stuff up.

If Meridian are swallowing the costs, then they're showing real commitment to their customers. It would be a lot cheaper to stick standard products in there.

Having said that - I've never seen inside their DVD players or processors, so I don't know how much FPGA is in there. Perhaps as you say they're working on a new one?

Good point, ultima. I think you are right on the FPGA vs ASIC issue.

Good post Mark. I'll reiterate though that just because it has a Meridian badge on it doesn't mean it's state-of-the-art.

I've been hanging around the video processing sections of AVS and AVForums for a while now and can say that neither Meridian nor Faroudja are considered serious players when it comes to the latest video processing. That may change.

The following comment may be a little harsh, but being the world leader in psycho-acoustic DSP audio does not make you an expert in video processing. Reputation and consistent delivery is everything in this game, which is why I have just committed to Lumagen for the foreseeable future. The future is bright. Radiant in fact.

Bob Stuart says alot of things, many of which are not true. If we'd all listen to Bob, we'd stick to DVD for another couple of years and forget about HD. :rolleyes:

FPGAs are very common, if you look hard enough, you'll probably find coffee machines that have FPGAs and are fully programmable. That doesn't mean you can use them as a phone...

Of course all videoprocessors are programmable, but not fully as Bob wants us to believe. Their current hardware is very limited, but of course Bob wants to sell his stuff and I can't blame him for making claims that are not 100% accurate. Everyone does, just look at Microsoft and the piece of junk they call an operating system.

But again, with their current hardware, they will not implement certain features. Some features could be done, even with current hardware. Why they do not give us all these features, I don't know. But then again, it's Meridian pissing off (potential)customers, not me.


FPGAs are universal and probably cheaper (depends on the design), but way less efficient. It's a question of what you'd want to do with it. You don't need TFlops of performance for a cell phone or coffee machine...

The MRI example is a little flawed isn't it? All the image processing is done in software on worksations. That's where the performance is. Play a little with Syngo on older and newer workstations and see how that turns out. Big reduction in scanning time, even compared to the old Numaris.

Hi Stephan,

We're talking performance an order of magnitude above what it possible in software, but it depends what you want them for. There are $1 FPGAs and $4000 FPGAs. The stuff that's doing real time predistortion and beamforming is high end, and your coffee machine is probably using a cost reduced mainstream product, more likely it'll have simple PLDs instead of FPGAs. The point is that FPGAs give access to massive performance without the requirement of huge production runs. The cost is per-item.

The programmability obviously only extends to the digital domain, so your coffee machine phone is unlikely , but modern FPGAs really are a blank slate for ones and zeros.

Having said that, I'd admit that TFLOPS aren't an FPGA thing. Floating point is hard in hardware. But integer maths just rocket, in MIPS they're incredible. Take a look at these:

http://www.newscientist.com/article.ns?id=dn7448

http://www.theregister.co.uk/2001/03/29/nasas_new_supercomp_sits/

http://www.electronicstalk.com/news/nal/nal130.html

Three supercomputers that use FPGA as the processing fabric. They're just the first ones I got from googling FPGA and supercomputer.

Your argument about what FPGA is actually in the box is a strong one - like the guy in Mad Max says; speed is just a question of money - how fast do you want to go? An 8051 is a microprocessor, but it's unlikely to worry your Xeon. Same applies to FPGA.

Then again, to stick FPGAs in a consumer product, and actually use the configurability is pretty new - up till now it's just been the big rack mounted stuff.

My argument is for the technology rather than the implementation, I think Meridian are cutting edge to be doing this in a consumer product.

The nearest parallel I could draw is 3G mobile. This requires huge processing in the base station. The spec took a long time to settle, too. Virtually all the manufacturers used FPGAs to give them performance and flexibility until things settled down. And now this is rippling down.

I think I've veered to far off topic again - sorry.

Cheers,
Mark.

I think one other thing that factors in here is some of these decoding/encoding protocols for copy protection etc. have to be used in order to get licensed and are not FPGA based and are not allowed to be.

Blimey....I'm not sure if bebop needs the lawyer or someone else.

:rolleyes: