Converintg analog stage to balanced on dual diff dacs? check pic.

[kasra]

I have a rotel rcd-975 with dual differential dacs, and single ended outputs.

I have sketched the analogstage of the player for one channel.
The d/a converter is tda1305:



I want to create a new balanced analogstage, and im wondering if thats possbile?




/kasra

[jwb]

The TDA1305 is a stereo DAC with a single-ended voltage output for each channel. Perhaps Rotel has rigged it so that the DAC for the left channel gets a digital datastream with the inverted and noninverted signals substituted for the left and right channels? I would classify that as a hack, but whatever.

The TDA1305 is highly integrated, so the outputs require no current-to-voltage conversion and no low-pass filtering. The output is at line levels, so you could simply take the two voltage outputs directly to an XLR connector.

[kasra]

Hello JWB.

Thank you for the reply.

There are a couple of other cdp out there with tda1305 that has balanced outputs such as Cambridge cd6 and Pink triangle something.

If i connect the signals directly to the output, doesnt the output level decrease ? What advantages/disadvantages does the outputstage give me?
I know others have done what you described, but they used outputcapacitors.

If i want to stick to an outputstage can i use the same method as this picture i found on the forum?



regards - kasra

[Circlotron]

Kasra, I posted that little opamp circuit a while back but I have a feeling that it tends to go hard over to one rail. I can't really remember now. Better you throw it together and see if it actually works before you do anything much with it.

GP.

[ultranalog]

I want to create a new balanced analogstage, and im wondering if thats possbile?

That's very possible. The TDA1305 is a single ended dac but Rotel used 2 per channel, yielding a nice differential signal.

Since every dac has 1.5V output, the differential output is 3V. It is quite free of spurious components:



Since the dac is a 3rd order sigma/delta type with 6 bit output stage, it's easy to explain why the spurious components are so low.

You can go several ways:

1. drive the output from the dac directly. I think if you have a high-Z (pre)amp, there's no problem. Just apply a coupling capacitor. Or make the output balanced if you have a balanced (pre)amp. You can go DC then.

2. use a 2:1 transformer to step down the voltage. You also achive a fourfold output impedance reduction and DC offset is also taken care of. The transformer also blocks any residual high-f stuff.

3. use an active stage. I did the latter (with tubes and 100kHz BW stepdown transformers - http://www.ultranalog.com/cdenhancer/cdenhancer2.html).

You'll find the sound of the TDA1305 is excellent ! It seems this dac was never meant to be 'high-end' (thd is at -80 dB SE or -90 dB differential, even below 16 bit resolution spec) but the topology is quite clever and allows for a simple analog stage.

Good luck modding,
Remco

[dice45]

Remco,

how dod your option 3 turn out to be sonically? I know 2 completely homebrewn DACs which use a tube output stage with a step down output transformer ... both have truly amazing sonics.

[ultranalog]

Originally posted by dice45
Remco,

how dod your option 3 turn out to be sonically? I know 2 completely homebrewn DACs which use a tube output stage with a step down output transformer ... both have truly amazing sonics.

The CD enhancers are after a year and a half and more than 20 produced (and some by others as well) still the best I can do. All I have been able to better is details and components. I have a test version running from the output of an SACD player with 6BX7's helped by transistor CCS'es feeding into an amorphous core OPT. Output impedance is 8 ohms (!). Distortion below that of the dac. All this is is a super-duper version of the same design, nothing new. The sound of the output stage I would describe as transparent. You retain the 'signature' of the dac used (philips=warm, bassy, pioneer=relaxed, neutral, burr-brown=analytic, holographic, sony=dynamic, transparent, etc).

What fascinates me about the sigma/delta dacs, is that most of them are voltage output types, with only one bit weight. This avoids the necessity of I/V conversion, with the inherent jitter susceptibility in that. Jitter susceptibility is greatly reduced i any case by the randomizing nature of the encoder used. If the dac utilizes a good way of smearing its asymmetry, as e.g. Pioneer and Sony do, you are virtually jitter immune (Guido's klok did nothing vastly different in my Pioneer player).
Furthermore, the in-band noise-freeness and (often available) differential nature allow for a simple analog stage with all functionality included. IMO simpler is only good if it still has all functionality.

Since I've built so many of them, I've also done a lot of measurements. A lot. A huge lot... Although I'm an audio nerd, I am also a science nerd and while I like it when something is better, I can only truly appreciate it if I know why (this is IMO the only way to progress). Correlating audible results with measurements is very satisfying. My preliminary conclusion is that it's the absence of (T)IM distortion that gives a great advantage over opamps. Especially with 1 bit outputs (every bit is a full-scale transient), opamps cannot handle the steepness of the transients and run without fb half of the time, thus making for a two-faced circuit: slewing while without feedback, and being highly linear when the feedback kicks in. The distortion created while slewing is a Fourier product of both audio as well as sampling frequency as well as the duraton of the slewing period, which in turn leads to unpredictable products folding back into the baseband. According to propability theory, this produces a normally distributed layer of noise over the audio (with amplitude increasing as frequency increases (!)), none of which is measured when doing the normal sine wave testing. I have done some modelling and come up with a way to actually measure the consequences of IM distortion with arbitrary input signals (and not just sines, sines say next to nothing). The differences are dramatic. Although the measuring method does not provide much insight into the distortion mechanism, it clearly shows the magnitude (and I already know the cause :P). I can't describe this method into much detail yet, since I've made it part of my university research in order to combine work and pleasure. Hope to have the theory complete in a few months (along with some less expensive solutions than amorphous core OPTs...)

Regards,
Remco

[janneman]

Originally posted by ultranalog


[snip]
I have done some modelling and come up with a way to actually measure the consequences of IM distortion with arbitrary input signals (and not just sines, sines say next to nothing). The differences are dramatic. Although the measuring method does not provide much insight into the distortion mechanism, it clearly shows the magnitude (and I already know the cause :P). I can't describe this method into much detail yet, since I've made it part of my university research in order to combine work and pleasure. Hope to have the theory complete in a few months (along with some less expensive solutions than amorphous core OPTs...)

Regards,
Remco

This is very interesting, can't wait to get the details. Are you sure you can't just give us a hint...?

Jan Didden

[ultranalog]

Are you sure you can't just give us a hint...?

The basic idea is to measure the cross-correlation power spectrum between dac_out and outputstage, taking into account the amplitude distribution of the signal applied, since this indirectly determines the spectral distribution of the HF remains in a sigma/delta coder. The latter is complicating the case significantly, because this also involves the time domain again. It has always been assumed the quantization error is distributed uniformly (a remainder of the linear PCM thinking while sigma/delta is not linear and can only be modeled using propability and signal theory), when in fact it's not, which greatly affects the ratio between signal and quantization error spectral powers and hence the signal to noise ratio. This holds in my models for any coder order and is only attenuated (but not changed in shape!) when multi-level coding is applied.

I hope my previous statement didn't come across as an attempt to protect IP, it's just so that at the moment I'm buried in theory, models, equations and measurements. I'm having a hard time wrapping my brain around it myself and am as of yet distrustful of the margin of error in both models and measurements. Until I know for sure what I'm doing, I really can't advise anybody else.

Regards,
Remco

[kasra]

Ultraananalog:
Thank you for your posts.

Have you heard of any success stories when using cdenhancer with tda1305?

I must admit i have always had your cd enhancer in mind, but wasnt really shure, since im building a dac i wasnt supposed to spend much cash on the rotel.

I guess the only expensive part is the Sowters, have you got any good source for them?

best regards