noise performance of DAC855x DAC856x DACs

Vadim Vasilev

Other Parts Discussed in Thread: DAC8562, DAC8564, DAC8568, DAC8550, DAC8560, REF5050, DAC80004, DAC8814, OPA227

Hi All.

I'm searching for low power precision DAC with good stability, low glitch and low noise performance. I've found that DAC855x DAC856x family is quite what I need. However while examining personal datasheets several question appeared:

1) All named DAC consists of upper divider resistor and lower string section with same total resistance followed by fixed gain 2 amplifier. Value of upper resistor (and total string resistance) for all models except DAC8562/3 is 62 kOhm. For DAC8562/3 value is 170 kOhm. Gain 2 amplifier in all DACs except DAC8562/3 feedback is formed by two equal resistors valued 50 kOhm while in DAC8562/3 by 150 kOhm resistors.

Is it true that DAC8562/3 resistors value are different from other family member ones?

2) Values for NSD at zero scale for DAC8562/3 and some other named chips looks too optimistic. Gain 2 amplifier having 150 kOhm valued resistor in feedback could not produce NSD lower than 70 nV/√Hz while figures 45 and 46 of datasheet provide its value dropped below 15 nV/√Hz. Lower noise may occur if amplifier is in saturation region but in this case NSD data has no sense and is invalid. Mid scale and Full scale NSD at fig. 45 look adequate at frequencies higher 1/f corner: estimation gives resistors noise at output 95 and 101 nV/√Hz while figure provide values of 102 and 111 nV/√Hz. Such too optimistic ZS NSD also found in other datasheets (DAC8564/5). DAC8568 include more realistic NSD data for zero scale with 90 nV/√Hz at 1kHz.

Is Zero Scale NSD in (DAC8562/3/4/5) datasheets accurate or not?

3) DAC8562/3 NSD data go to 1000 nV/√Hz at 10 Hz. While having 1uV of noise density at 10Hz one could not obtain 2.5 uV p-p noise in 0.1..10Hz bandwidth - even in case of white noise it should be 20 uVp-p. So 1/f corner at NSD figure looks not really existing and NSD value is suggested to be below 200 nV/√Hz at 10 Hz if 2.5uVp-p value in 0,1..10Hz bandwidth is true.

Is NSD at 10 Hz really below 200 nV/√Hz or is LF noise significantly higher than 2.5 uVp-p?

4) DAC8562/3 NSD for Full Scale code with external reference applied is according to figure 45 approximately 110 nV/√Hz. DAC8550/2/4/5 have 3 times lower valued resistors in string/buffer section so DAC noise is suggested to be lower than for DAC8562/3. However DAC8550/2/4/5 datasheets have no NSD and LF noise data in characteristics section but NSD obtained from typical characteristics figure value is 170 nV/√Hz at 1kHz for all chips. I suggest that this is not characteristic of DAC chips but of some unnamed reference used in setup.

In this case what is real noise performance of DAC8550/2/4/5 with external reference noise excluded?

5) In my assumption I suppose that all named chips (DAC8550/2/4/5, DAC8560/2/3/4/5/8) have same converter unit and differs only by presence or absence of internal reference (855x vs 856x), number of channels (1/2/4/8), possibility of power-on reset to ZS or MS and to select gain=1 or 2. If it is true I suppose having equal analog performance in equal condition (same external reference, power etc). However datasheets characteristic vary between chips. This apply not only to noise characteristics I described before but to glitch performance, thermal drifts and so on.

Could I expect equal analog performance (INL does not matter; only thermal and long-term stability, noise, glitch taken into account) from any chip from list (DAC8550/2/4/5, DAC8560/2/3/4/5/8) or does some chips from this list have optimized design that really gives better parameters (stability, noise, glitch)?

Best regards.

Vadim.

over 8 years ago

0 Jon Key over 8 years ago

TI__Intellectual 920 points

Hi Vadim,

Thank you for posting here on our forum. I took a look at your questions today and have the following responses.

1.) Is it true that DAC8562/63 resistors value are different from other family member ones?

Yes, though all of these devices are string DACs there are sub-variants within the string architecture whose differences can contribute to additional parametric distinctions. The DAC8562/63 have a different architecture than all the other DACs you mentioned. In fact DAC8560 has a different architecture from DAC8564 and DAC8568 as well.

2.) Is Zero Scale NSD in (DAC8562/3/4/5) datasheets accurate or not?

Your observation is true that the output amplifier has 150 k Ohm feedback resistors which would produce a NSD of 70 nV/rtHz. However, in these different architectures the DAC amplifier output stage has little to no gain at zero scale which is why the noise is so low.

3.) Is NSD at 10 Hz really below 200 nV/√Hz or is LF noise significantly higher than 2.5 uVp-p?

For low frequency ranges (0.1Hz to 10Hz), the peak-peak noise provides a better picture of the noise performance of the DAC since the spectrum analyzers margin of error increases as we go lower in the frequency. Therefore, 2.5 uVpp would be a good estimate for the flicker noise specification of this DAC.

4.) In this case what is real noise performance of DAC8550/2/4/5 with external reference noise excluded?

Since these DACs need an external reference to function correctly, you will always have reference noise added to the overall noise spec. However, theoretically, one can calculate the DAC noise contribution by taking out the reference contribution. We typically use low noise reference such as REF5050 to be used with our DACs. Note that the reference noise and DAC noise will be uncorrelated; therefore RSS technique must be used for the analysis.

5.) Could I expect equal analog performance (INL does not matter; only thermal and long-term stability, noise, glitch taken into account) from any chip from list (DAC8550/2/4/5, DAC8560/2/3/4/5/8) or does some chips from this list have optimized design that really gives better parameters (stability, noise, glitch)?

As mentioned previously, all of the DACs you are considering have different architectures so they will not necessarily have the same analog performance.

If you could share what design goals you are looking for, specifically with regards to noise, INL, total unadjusted error, glitch etc. I would be more than happy to recommend some parts to meet your needs.

Thank you,
Jon

0 Vadim Vasilev over 8 years ago in reply to Jon Key

Prodigy 70 points

Hi, Jon.

Thank you for useful and detailed response.

However I cannot agree with you in all details.

1. I looked DAC8560 and DAC8564 datasheet and cannot find differences in "architecture" that may lead to different noise and stability performance: string resistance is same, buffer feedback resistance too. Please explain what is the difference between architecture of analog circuitry of DAC8560 and DAC8564 (and DAC8568) that gives best analog performance (noise, drift, crosstalk) to 8-channel product and worst to single channel product while their DAC architecture descriptions are identical and values that determine noise characteristics are same.

2. You wrote about DAC8562 that

Jon Key said:
in these different architectures the DAC amplifier output stage has little to no gain at zero scale which is why the noise is so low.

I still insist that gain=2 amplifier with 150kOhms resistors (shown in figure 88 of DAC8562/3 datasheet) could not produce lower than 70nV/rtHz of device noise just because of resistors thermal noise. The only possible reason for such a amplifier to demonstrate lower noise value at figures 45 and 46 is if its output is out of linear operation mode. I understand that term "Zero scale" suppose that code 0x0000 is used but because of non-linear operation of internal circuitry characteristics obtained with this code are out of practical use and sense. From practical point of view "Zero scale" characteristics (noise, temperature dependence) should be obtained with code that is small enough but that guaranteed shifts DAC analog circuitry to linear region. It is already done so for "relative accuracy" characteristic in datasheet and this is reasonable. Lower and upper 512 codes of 65536 range where adequate DAC operation is not guaranteed should be excluded from parameterization of noise and stability related characteristics. The only "Zero scale" characteristic that should be measured with true 0x0000 code is Zero-code error that is analog of opamp rail headroom in this case.

3. Ok. Thus DAC NSD is lower then 1000nV/rtHz at 10 Hz and this is good.

4. When DAC noise density and low-frequency value is provided in datasheet I assume that it is rule of thumb to use intrinsically low-noise reference when we may ignore its (reference) noise, when manufacturer compute DAC related noise based on (simultaneously sampled) reference noise and reference+DAC output noise (best cases) or just inform user about setup used when those characteristics are obtained.

Thank You for details about reference used. Unfortunately REF5050 datasheet also lack from noise specification: Electrical Characteristics section provide only LF noise of 3uVp-p/V (!) and there is no NSD data. NSD details may be found in https://e2e.ti.com/support/other_analog/etc_analog/f/247/p/376073/1328580 thread

Then according to datasheet and blog data REF5050 should probably have 15uVp-p of LF noise and 340 to 800nV/rtHz NSD at output depending upon which noise reduction technique is used. This data match 170nV/rtHz at 1 kHz obtained at output of DAC855x DAC with midscale code loaded. This means that DAC intrinsic noise is much lower than reference used (REF5050) noise and Output Noise Density figure in DAC855x datasheets provide not DAC noise but reference used one.

5. same as 2. - What are differences in architecture? Please explain.

My system is open/closed loop control system for physical setup (several channels with different requirements). The are requirements to low power consumption, good resolution (1e-5), low noise, low glitches, excellent stability (both thermal and long-term, same as resolution) and definitely affordable price. INL and TUE are not of great importance to me.

Thank you.

Best regards.

Vadim

0 Jon Key over 8 years ago in reply to Vadim Vasilev

TI__Intellectual 920 points

Hi Vadim,

I’m sorry that my last post didn’t clear up these issues.

1.) DAC8560, DAC8564, and DAC8568 are string DACs that have the same block diagram in the datasheet, however, as mentioned in the previous post they are sub-variants of the string DAC architecture and have different output amplifier architectures as well. The block diagram is a simplified version of the DAC architecture, in practice we employ multiple design and layout techniques, on a high level this may include segmentation techniques.

2.) You wrote:

Vadim Vasilev said:

The only possible reason for such an amplifier to demonstrate lower noise value at figures 45 and 46 is if its output is out of linear operation mode.

This assumption is correct and the reason the noise is so low at zero scale. On this device we measure the zero scale NSD with code 0x0000 loaded into the DAC register.

Your observation about measuring noise within the linear operating region of the DAC is something we have heard from many customers. On datasheets for newer parts, such as DAC80004, we have begun measuring the NSD from code 512.

3.) Excellent, glad I could help clear that up.

4.) Your assumption is correct that the datasheet is reflecting the reference noise more than the noise from the DAC, but since we are unable to operate the DAC without a reference this is the data we have in displayed in the datasheet.

5.) Again as mentioned above I am unable to discuss this in detail as this pertains to TI’s IP.

Vadim Vasilev said:

The are requirements to low power consumption, good resolution (1e-5), low noise, low glitches, excellent stability (both thermal and long-term, same as resolution) and definitely affordable price. INL and TUE are not of great importance to me.

If you are concerned with low noise and low glitch I would recommend looking into our line of MDACs, these DACs are typically unbuffered which will allow you to use an output buffer with much lower noise output than those included in our buffered string DACs. I would recommend looking into DAC8814 and then using a low noise op amp such as the OPA227 you would receive much better noise and glitch performance than the string DACs mentioned above.

However, MDACs are typically more expensive than string DACs. If the cost is an issue with DAC8814, I would suggest looking into DAC80004 and pairing it with a low noise reference. DAC80004 is a quad channel 16-bit voltage output DAC that has a maximum output of 5 V.

Let me know if this helps, or if you have any more questions!

Thank you,

Jon

0 Vadim Vasilev over 8 years ago in reply to Jon Key

Prodigy 70 points

Hi Jon.
Thank you a lot.
I'm looking for string DAC as compromise solution.
MDAC usage leads to design complication and require more power and space: negative supply generation, separate (from other system's part) negative voltage reference, additional amplifiers. And probably cost is high enough. So this variant has definitive advantages but drawbacks too.
DAC80004 power consumption is higher than I planned to use.
Nevertheless, thank you for variants proposed. I will pay more attention to them.
Best regards.
Vadim.

Data converters

Data converters forum

noise performance of DAC855x DAC856x DACs