Building the ultimate NOS DAC using TDA1541A

I would recommend moving to a RAID setup if you are worried about losing data. With Raid, 2 or more drives are used together to make up a single logical drive.

If you use Raid 5. You have a very safe option for Data. if say you have 3 hard, 2 are used for storing data, the third for redundancy.

If one drive is to fail, the redundancy data can be used to rebuild the data from the failed drive onto its replacement. Very usefull.

The more drives you use the more economical the setup, as one redundancy drive is required.

i.e.
3*40gb Drive = 80gb
5*40gb Drive = 160gb

Hardware Raid is best, but Windows has software raid built into the modern operating systems. The disadvantage is that you have to have the raid array on different drives to the OS in the latter.
 
Ecdesigns,

Yes, write errors do occur, of course. I never said anything else. My point was that from my experience, write errors so severe that the ECC info is not sufficient to correct them seem to have been extremely rare, if occuring at all.

If data integrity is extremely important, there are further measures one can take. Disk redundancy in the form of RAID arrangements, as somone suggested, is probably the most common way to go. One could also test read every sector after writing it, but that would waste too much time on most systems, and is probably only used in systems with extreme requirements for reliability.

The bottom line is, of course, that write errors on hard disks is probably so rare that we can forget about them if the purpose was to store audio.

The second bottom line is, disk crashes happen and cause a lot more headache than one or a few write errors would, so always back up important data regularly.
 
Administrator
Joined 2004
Paid Member
Hi everyone,
Storing and playing a music collection would be expected to last for many years. The investment in your music files and time to rebuild them would be enormous.

I personally do not recommend RAID5 setups for this. Most people could not set up the drive farm properly from a power supply / cooling standpoint. The cost is yet another issue.

What would I recommend? RAID2 using JFS (Journaled File System). This is only drive mirroring and it's very effective. I've run systems like this and this has saved my behind more than once. This was available with OS/2 Advanced Server 5 many years ago. A Linux distribution would be the most reliable for this purpose these days. Linux would also afford the very best virus protection with it's file permissions structure. Windows is just now getting to a point where others were over 10 years ago. With the built in ability to snoop your files that Microsoft always includes in their OS, they will always be less secure.

That's my take on it, for what it's worth.

-Chris
 
anatech said:
Hi everyone,
Storing and playing a music collection would be expected to last for many years. The investment in your music files and time to rebuild them would be enormous.

I personally do not recommend RAID5 setups for this. Most people could not set up the drive farm properly from a power supply / cooling standpoint. The cost is yet another issue.

What would I recommend? RAID2 using JFS (Journaled File System). This is only drive mirroring and it's very effective. I've run systems like this and this has saved my behind more than once. This was available with OS/2 Advanced Server 5 many years ago. A Linux distribution would be the most reliable for this purpose these days. Linux would also afford the very best virus protection with it's file permissions structure. Windows is just now getting to a point where others were over 10 years ago. With the built in ability to snoop your files that Microsoft always includes in their OS, they will always be less secure.

That's my take on it, for what it's worth.

-Chris

I dont think that raid 5 is that hard to set up effectively, however mirroring is effective as well. As far as setting up a moderate sized drive farm.... Its not rocket science, HD's ar not power hungry enough to defeat even modestly equiped PSU's and a medium quality or higher PC case will have plenty of drive bays, and a reasonably intelligent (though not optimum) cooling arrangement.

Mirroring is certainly simple enough, though it is more expensive per GB once you reach a certain size of data storage.

Anyway i suspect i am about to get my knuckles rapped for going too far OT.
 
Administrator
Joined 2004
Paid Member
Hi justblair,
Hey, it's "just me" here. ;)

Most people are unaware and unlikely to take all the steps needed. They are setting up a music server, not a workstation. So they will not invest in the proper equipment to do the job.

Drive mirroring is cheaper still. It runs on IDE or SATA, not just SCSI. Cheap and cheerful. Also, easier on the power supply on start up and the cooling than 5 bays of whining drives. I would tend to put the OS on a smaller boot drive, not on the array.

-Chris
 
nicoch46 said:
SanDisk SSD delivers an outstanding two million hour mean time to failure (MTTF)1 rate (is not pendrive!)
64gb.....not bat and this is only beginning


64 MB, yes for solid state hard "disks" that uses multiple flash chips. Of course one can use any number of flash chips to get the desired capcity. Price still has to drop to be competitive, but of course it will in a few years time.

The MTTF figure most probably does not refer to read- and write errors, but to device failures. I am in no way en expert on flash memories, but according to what I have read they do have a maximum number of writes (or rather erases) of between 10 000 and 1000 000 times per cell, which in practice means per block. Often algorithms are used to allocate free blocks for writing that minimizes the maximum number of writes to any block. But eventually some blocks will start to fail due to too many writes, and the available amount of space will decrease. This might not be a problem for most storage applications. It could be a problem if using the flash disk for swap space and other frequent-write applications.
 
Administrator
Joined 2004
Paid Member
Hi Christer,
The only thing I'll add to the excellent points you've made is that I've had enough SanDisk failures to be suspicious of the brand, and the technology. And those were used for data transfer (still babies when they died) and digital cameras (again, babies and DOA's). I'm not fan of these, but I would be if they were more reliable. We tried doing phone system programming backups with them. Now we when back to saving files on the laptop HD.

-Chris
 
Chris,

are you referring to Sandisk only, or flash technology in general? And do you mean device failures, that makes them unusable or unreliable, or do you mean they are unreliable as in having many read-/write errors?

I have far to little experience of the technology. The only one being the Sandisk card in my digital camera, which has worked well so far.
 
Administrator
Joined 2004
Paid Member
Hi Christer,
Complete failures. Cards that formatted fine and failed to capture data. More headaches than I care to mention. All with Sandisk. Some failures with the digital camera. Nikon and the cameras checked fine by Nikon. Other cards worked great.

I haven't bothered to try lately. I imagine they must have improved them. I know this because they are still in business. Could have been a bad batch I guess.

-Chris
 
I2S Reclocker - How Does It Work?

Hi ECDesigns,

I have added your reclocker http://www.diyaudio.com/forums/attachment.php?s=&postid=1025489&stamp=1160403522
to one of my single TDA1534 chip DACs. Thank you for sharing, it was a fun evening.

The DAC was built a week ago from all new parts and is still breaking in. I added a jumper/switch on the board so I can switch on/off the reclocker. The TDA1534 either has the BCK signal direct from the PCM2706/7 chip or from the reclocker.

There is a noticeable difference in sound between the two settings. Reclocked BCK is the setting I prefer, at times it seems to be missing something I enjoyed from the direct BCK setting. I likely need more time to come to grips with what I am hearing or not hearing. At a later date I will comment on how it affects the sound. I am trying to not let DIY building emotions affect my judment so much :)

The point of my posting is to ask you (or anyone who can) to help me understand the reclocking. From looking at the logic I see that the BCK signal from the PCM2706/7 is delayed by one 48MHz clock cycle while the DATA and WS signals are untouched.

To me this makes sense to allow the WS and DATA signals to settle out of the jitter zone and BCK is now clocked from the 48MHz clock which has less jitter then from the PCM2706/7.

But I don't understand the relationship between the 48MHz clock and BCK from the PCM2706/7. I wish I knew how the PCM produces BCK from the 12MHz clock. I found the SPACT patent but it is too broad in scope to tell me much http://www.google.com/patents?id=8tgRAAAAEBAJ&printsec=drawing&zoom=4&dq=spact#PPA26,M1 .

If 48MHz was evenly divisible by BCK and synchronous to it then it would make perfect sense to me. But since they don't seem to be then how can they act together to produce a correct reclocked BCK signal without beating or at all?

Please enlighten me, this has been bugging me for a week now, I can not get it out of my mind :)
 
Reclocking

Hi Brent Welke,

Thanks for your reply [post # 1595]


The point of my posting is to ask you (or anyone who can) to help me understand the reclocking. From looking at the logic I see that the BCK signal from the PCM2706/7 is delayed by one 48MHz clock cycle while the DATA and WS signals are untouched.

The shiftregister reclocker delays BCK by two clock cycles (U5: QB). The introduced BCK delay is only marginal, so there is no need to reclock both DATA and WS as well.

When using TDA1543 / TDA1541A, BCK needs to have low jitter, since BCK determines the exact moment the sample is placed on the DAC output.

The function of the (shiftregister) reclocker is to lock the fluctuating BCK clock signal to the master clock transients. BCK can now only change on the (positive) going edge of the master clock.


To me this makes sense to allow the WS and DATA signals to settle out of the jitter zone and BCK is now clocked from the 48MHz clock which has less jitter then from the PCM2706/7

Both DATA and WS are still sampled correctly, the only difference is that BCK jitter, introduced by the PCM2706 PLL / clock recovery circuit, is now removed by "locking" BCK to the low jitter masterclock transients.


But I don't understand the relationship between the 48MHz clock and BCK from the PCM2706/7. I wish I knew how the PCM produces BCK from the 12MHz clock. I found the SPACT patent but it is too broad in scope to tell me much

It's logical to use a master clock frequency that's close to a multiple of BCK, but in case of the PCM2706, we also need a 12 MHz clock. Since the PCM2706 PLL is influenced by the 12 MHz clock signal, it's not a bad idea to run both 12 MHz clock and the master clock in sync.

So after some calculating, I ended up with a 48 MHz master clock. When divided by 4, I get the desired 12 MHz clock signal for the PCM2706. The master clock is almost 17 times higher than BCK (48,000,000 / 2,822,400 = 17.0068).


If 48MHz was evenly divisible by BCK and synchronous to it then it would make perfect sense to me. But since they don't seem to be then how can they act together to produce a correct reclocked BCK signal without beating or at all?

Well the 48 MHz comes pretty close already (17.0068, deviation: 0.0068)

The custom made oscillators with the WF10192 chip has a specific 16 MHz crystal type that oscillates at a slightly lower frequency than 48 MHz (third overtone), in fact it's very close to 47.98 MHz. This results in a 11.995 MHz clock signal for the PCM2706 (within tolerances), and 47,980,000 / 2,822,400 = 16.9997 (deviation: 0.0003). So now the master clock is almost a perfect multiple of BCK as well. It's even possible to have custom crystals made that are suitable for operation at the third overtone (15.9936 MHz), when multiplied by 3, we get 47.9808 MHz, an exact multiple of BCK.

The "synchronous" part is a bit more difficult, since both BCK and the master clock aren't synced. Here's where the shiftregister reclocker comes-in. The shiftregister reclocker is a combined digital one-schot, and dual-stage asynchronous reclocker. 9 master clock pulses after the rising edge of the BCK input signal, both shiftregisters receive a forced-reset (I needed 2, in order to achieve a delay of 9 master clock pulses). Now the shiftregister reclocker is ready to get a pre-defined lock on the next rising edge of BCK (since it's reset slightly earlier). This mechanism enables BCK to be synced to the master clock.
 
TDA1541A DEM clock frequency / decoupling cap value

Hi onnosr,


Thanks for your reply [post #1596]


I tried the dem osc running on 705,6 khz (instead of 352,8)
coupled direct via 470 pf to pin 16 (pin 17 0.2 mf to agn)

soundstage is more massive imho

Pse give it a try and tell us your outcome.

the tda1541a is able to handle that freq.


The TDA1541A active dividers have their limitations (maximum switching speed), pushing these limits could result in increased errors (settling time), and increased power consumption. TDA1541A chips have some tolerances, so one chip could work fine with 705.6 KHz, the other wouldn't.

However, there seems to be a consistent improvement in perceived sound quality with increased DEM clock frequency. So the effect you observed is correct.

The answer is simple, the filter (decoupling caps) produces lower current ripple as the DEM clock frequency increases. So instead of pushing the DEM clock frequency limits, it's also possible to increase decoupling capacitor values.

The DA1541A modules with the new smaller heatsink can accomodate 0.68uF decoupling caps. This is possible by placing half of the caps on the component side, the other half on the solder side (there is plenty of room underneath the modules). The 1 uF for MSB could be increased to 2.2uF if desired.

That should be more than sufficient to get very low ripple current on all active divider outputs.
 
BCK Reclocking

Hi ecdesigns,

Thank you for answering my questions about your reclocking. I was glad and sad to hear I hadn't missed something simple :)

What minimum scope spec would you recommend to be able to view BCK jitter like the images you posted? Not that I can run out and buy a new scope, but who knows. Maybe some kind sole will invite me over to use his scope :xeye:

I have done more A/B testing between reclocking on/off. On the same DAC reclocking always sounds better. Until recently my favorite (burned-in) non reclocked DAC was more pleasant to listen to then the new reclocked DAC. Other then reclocking the DACs are identical in parts.

I would recommend all PCM2706/7 I2S users use/try your reclocker, even those with with super clocks.

To modify my DAC I used SMD parts to make the mod almost invisible. Here is my DigiKey.com parts list for anyone interested in a SMD mod for their existing USB DAC:

568-2619-5-ND IC SYNC 4BIT BINAR COUNTR 16SSOP
296-14020-1-ND IC SHFT REG SERIAL 8BIT 14-TSSOP
NC7SZ04P5XCT-ND IC INVERTER UHS SINGLE SC70-5
CTX315LVCT-ND OSC CLOCK 48.0000 MHZ 3.3V SMD

My camera is out of town for a week or I would post a picture of it.

~~~~~~~~~~~

Point-to-Point SMD building tip -> before you begin rub off all IC pins that are 'No Contact' or bridge them to other pins, which ever works best for that pin position. Buy extra chips. The more you work with SMD parts the easier it becomes. At first you think there is no way you could use such small parts.

~~~~~~~~~~~

USB A/B testing tip -> On a Windows machine using the latest versions of Foobar and ASIO you can send the same audio stream to more then one USB DAC at a time. This makes for very nice A/B testing and to burn-in many DACs from one machine.