Marce,
Can you give me some insight into why 12-14 layers, how does this number compare to a circuit which is not very signal intensive (i.e. from a connectivity point of view could easily be done on 2 layers).
Or of these 12-14 layers how many are needed for just trace routing, how many are planes.
I'm sure it's complicated with impedance control and cross-talk and low Z power distribution, etc but just some rough idea's would be helpful.
Thanks
-Antonio
Those old tube computers are a perfect example of the common misconception that digital is discrete (on or off) in practice, obviously the vacuum tube diodes used in the computers back then and the transistors used to transmit digital signals now are "analog" or continuous in the transition from what is defined as on to off. That doesn't mean that digital theory isn't practical just that its easy to fall into the trap of confusing a mathematical abstraction with physics
Discrete just doesn't exist in nature, its the Heisenberg Principle. Just think how much better off we would be had Sony/Phillips execs grasped that simple concept before sticking us with spdif for the last 30 years.
Oh well as of today we have Internet Explorer 10 and I get automatic spell check when posting here, so digital isn't all bad
Really I haven't played an analog tape or vinyl in 20 years, digital can be fantastic you just have to cheat
Discrete just doesn't exist in nature, its the Heisenberg Principle. Just think how much better off we would be had Sony/Phillips execs grasped that simple concept before sticking us with spdif for the last 30 years. Oh well as of today we have Internet Explorer 10 and I get automatic spell check when posting here, so digital isn't all bad
Really I haven't played an analog tape or vinyl in 20 years, digital can be fantastic you just have to cheat
Last edited:
Well, their predecessors are even better examples. They used relays. Relays, especially small signal ones, really are discrete - on or off.
Again, digital circuits use analog components in a digital way, to transmit and process discrete, binary signals. That is done by requantizing, resyncronizing and regenerating the "analog" signal so that it becomes digital again - truly just on or off.
Digital circuits, at the level of digital circuits, are really discrete and purely digital.
Talking about things lie s/pdif is confusing the issue, because s/pdif isn't a digital circuit, it is a transmission system. Transmission systems are, on the level of the physical media, analog, but again, get re-digitized in the receiving circuitry.
Vacuum tube computers are no more and no less analogue than a BJT-based computer or a CMOS-based computer. Provided timing is not an issue and the computer is well-designed then it really is digital, as far as the logic goes.
Digital audio interconnects are a separate issue, as there is a timing aspect too.
Digital audio interconnects are a separate issue, as there is a timing aspect too.
I guess we can keep repeating it until we are blue in the face. Unfortunately "digital is really analog" is such a nice soundbite for the "everything matters" crowd... Who cares if it is true or not, it *sounds* good...
Implying there is a "not everything matters" crowd ... hmmm, only have to get half the issues sorted out to get fantastic sound; plenty of time left then to sink a few beers, that mightn't be too bad, might hop across there for a while ...
Frank
Absolutely - I am definitely in the "if I really can't hear any difference and measure any difference, why should I spend a lot of money just because some 'audiophile reviewer' who gets paid to promote products tells me I should?" camp.
Hi Magnoman,
a recent 14 layer design had the following layer stack:
Top: Top layer components, GND copper pour, some routes
L2: critical signals
L3: gnd
L4: critical signal
L5: gnd
L6: power
L7: signals, some power islands
L8: signals, some power islands
L9: power
L10: gnd
L11: critical signal
L12: gnd
L13: critical signal
bOTTOM: bOTTOM LAYER COMPONENTS, gnd COPPER pOUR, SOME ROUTES
That was a standard PCB, through hole vias from top to bottom, a similar design done using HDI design had less layers due to the extra room this sort of design allows, this had vias top-L1, L1-L2, L2-L3, L3-L7, L7-L8, 8-L9, L9-bottom:
Top components, GND plane
L2 power
L3 critical signals
L4 gnd
L5 signal
L6 signal
L7 gnd
L8 critical signals
L9 power
BOTTOM
the main thing is to run critical signals (high speed, DDR interface, clocks etc) next to a contigous GND or power plane. To do the job properly we may spend 2 or 3 days just entering the constraints into the system, this gives us skew groups (delays between groups of signals), permited layers for certain signals, target impedance etc etc.
There are numerous variations on these builds, and often depending on system rerquirements and costs gnd planes can be added or reduced. A lot of PC cards can be layed out on six layers, with a buit of a struggle as the DDR interface is optomised by the interface IC's for DDR memory modules (the lines are all pretty and sorted out).
Another difference with todays boards compared to the PTH 0.1" dil based designs of yesteryear is routing style, the X:Y routing is not as prevelant, what you see these days is more river routing (one term for it, its Sunday morning so I'm a bit slower than normal, more like Quaps pet than human! today, yeast poisoning). The picture below is an example of todays routing, ie we try to minimise layer changes (unlike manhatton routing) so the traces meander between the vias of the BGA.
Marc
a recent 14 layer design had the following layer stack:
Top: Top layer components, GND copper pour, some routes
L2: critical signals
L3: gnd
L4: critical signal
L5: gnd
L6: power
L7: signals, some power islands
L8: signals, some power islands
L9: power
L10: gnd
L11: critical signal
L12: gnd
L13: critical signal
bOTTOM: bOTTOM LAYER COMPONENTS, gnd COPPER pOUR, SOME ROUTES
That was a standard PCB, through hole vias from top to bottom, a similar design done using HDI design had less layers due to the extra room this sort of design allows, this had vias top-L1, L1-L2, L2-L3, L3-L7, L7-L8, 8-L9, L9-bottom:
Top components, GND plane
L2 power
L3 critical signals
L4 gnd
L5 signal
L6 signal
L7 gnd
L8 critical signals
L9 power
BOTTOM
the main thing is to run critical signals (high speed, DDR interface, clocks etc) next to a contigous GND or power plane. To do the job properly we may spend 2 or 3 days just entering the constraints into the system, this gives us skew groups (delays between groups of signals), permited layers for certain signals, target impedance etc etc.
There are numerous variations on these builds, and often depending on system rerquirements and costs gnd planes can be added or reduced. A lot of PC cards can be layed out on six layers, with a buit of a struggle as the DDR interface is optomised by the interface IC's for DDR memory modules (the lines are all pretty and sorted out).
Another difference with todays boards compared to the PTH 0.1" dil based designs of yesteryear is routing style, the X:Y routing is not as prevelant, what you see these days is more river routing (one term for it, its Sunday morning so I'm a bit slower than normal, more like Quaps pet than human! today, yeast poisoning). The picture below is an example of todays routing, ie we try to minimise layer changes (unlike manhatton routing) so the traces meander between the vias of the BGA.
MarcAttachments
Sometimes my system and the sounds coming from it are not up to scratch and I can find them irritating, other times they are beautiful, audio nivarna, I have found that the main cause of this is my general mood and state of mind (though it could be that when cleaning the missis has disturbed the critical positioning of my wires, I like the spagetti look, cos thats how they end up), so I dont look for gremlins in my hardware, I tend to find somthing to smooth my fevered brow, and that may just be putting some gentle tune on that draws me gently into the music and relaxes my stress or
, and voila it all sound good again
), so I dont look for gremlins in my hardware, I tend to find somthing to smooth my fevered brow, and that may just be putting some gentle tune on that draws me gently into the music and relaxes my stress or, and voila it all sound good againHi Magnoman,
a recent 14 layer design had the following layer stack:
Top: Top layer components, GND copper pour, some routes
L2: critical signals
L3: gnd
L4: critical signal
L5: gnd
L6: power
L7: signals, some power islands
L8: signals, some power islands
L9: power
L10: gnd
L11: critical signal
L12: gnd
L13: critical signal
bOTTOM: bOTTOM LAYER COMPONENTS, gnd COPPER pOUR, SOME ROUTES
That was a standard PCB, through hole vias from top to bottom, a similar design done using HDI design had less layers due to the extra room this sort of design allows, this had vias top-L1, L1-L2, L2-L3, L3-L7, L7-L8, 8-L9, L9-bottom:
Top components, GND plane
L2 power
L3 critical signals
L4 gnd
L5 signal
L6 signal
L7 gnd
L8 critical signals
L9 power
BOTTOM
the main thing is to run critical signals (high speed, DDR interface, clocks etc) next to a contigous GND or power plane. To do the job properly we may spend 2 or 3 days just entering the constraints into the system, this gives us skew groups (delays between groups of signals), permited layers for certain signals, target impedance etc etc.
There are numerous variations on these builds, and often depending on system rerquirements and costs gnd planes can be added or reduced. A lot of PC cards can be layed out on six layers, with a buit of a struggle as the DDR interface is optomised by the interface IC's for DDR memory modules (the lines are all pretty and sorted out).
Another difference with todays boards compared to the PTH 0.1" dil based designs of yesteryear is routing style, the X:Y routing is not as prevelant, what you see these days is more river routing (one term for it, its Sunday morning so I'm a bit slower than normal, more like Quaps pet than human! today, yeast poisoning). The picture below is an example of todays routing, ie we try to minimise layer changes (unlike manhatton routing) so the traces meander between the vias of the BGA.
This is why people like NOS DAC's, what you describe is a far cry from diy-audio unfortunately. Unless that is you would like to setup a user programmable oversampling fifi pll digital filter pcb group-buy for the rest of us?
Your correct, this is the more extreme end of PCB design, where cost is secondary to reliability, signal integrity and low EMC emmisions. These layer counts can go up depending on the number of pins on any BGA devices, its that crazy that there is even a book on BGA breakout...
BGA Book Download
not the most exciting of reads, unless your faced with a 1000 plus pin device on 0.8mm pitch or smaller.
The biggest problem for DIY at these levels is cost, as well as multi layer they are generaly impedance controlled builds that adds to the cost, and then there is assembly, they are not the sort of boards you can assemble at home, with all the copper layers controlling the reflow profile requires the right kit and a multi zone oven, and all are double sided placement, it is the only way to get the decouplers in the correct place. Prototypes can be a few thousand a panel, production runs depend on volume and location of manufacture.
BGA Book Download
not the most exciting of reads, unless your faced with a 1000 plus pin device on 0.8mm pitch or smaller.
The biggest problem for DIY at these levels is cost, as well as multi layer they are generaly impedance controlled builds that adds to the cost, and then there is assembly, they are not the sort of boards you can assemble at home, with all the copper layers controlling the reflow profile requires the right kit and a multi zone oven, and all are double sided placement, it is the only way to get the decouplers in the correct place. Prototypes can be a few thousand a panel, production runs depend on volume and location of manufacture.
What I did hope for in this thread was something like 1 box, maybe single physical board (probably multi electrical domains) , USB -> speaker (spdif - speaker) solution probably with class-d output stage to reduce the number of d->a stages to almost zero (what is really 16 bit -> pwm sawtooth conversion?) sonically superior to any multi box, multi board (e.g. Buffalo card house) integration challenges. Room correction/ EQ - yepp. One streamlined clock distribution, physically decoupled and isolated. Buss bar power distribution architecture with proper regulation topology depending on context. 2 chennel but with expansion possibility provided by digi out.... Some disruptive thinking in mechanical packaging wouldn't hurt either
Think Devialet ++ ....
Didn't happen...
/
Think Devialet ++ ....
Didn't happen...
/
Last edited:
Is going from PCM (be it 16 or 24 bit) to pulse width modulation really any more direct than going from voltage proportional to a 16 or 24 bit PCM code to a pulse width proportional to that voltage?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.