TLV320ADC3140: VREF inaccuracy and Input range span in DC-coupled single-ended input.

Part Number: TLV320ADC3140
Other Parts Discussed in Thread: ADC3140EVM-PDK

Hi TI

We have a TLV320ADC3140 running on 3V3. AREG is measured to 1.8V and DREG is measured to 1.5V.

Question 1:

The TLV320ADC3140 is configured to MICBIAS=VREF=2.75V. 2u2 on VREF. But I measure it to 2.4V. There is a 20kohm load on MICBIAS.

If I change VREF to 2.5V, I measure it to 2.2V. If I change VREF to 1.375, I measure it to 1.2V.

I cannot figure out why. Please help.

Question 2:

A related question is the input range. The inputs are configured to DC-coupled and single-ended input. All gains at 0dB. No filters enabled. Using TDM 32-bit mode.

It seems that the valid input voltage range is from apx 0V to VREF/2 corresponding to digital output values from 0x0000 0000 to 0x7FFF FFFF. In pracis a bit less.

I was expecting a range from 0V to VREF with a digital output from 0 to 0xFFFF FFFF or alternatively -0x8000 0000 to 0x7FFF FFFF. But apparently not. But why.

E.g. a AC signal on 2.75V peak-peak with a DC bias at 1.375V to give approximately the largest possible signal to convert. I know that in practice the rails are not perfect.

It seems that either there is a gain on +6dB on the analogue input on chip, or alternatively the inputs voltage range is from -VREF/2 to +VREF/2. (I now that the latter is not very likely as there is no negative supply on chip, and the absolute maximum ratings specify the inputs not to drop below -0.3V)

The input range and the output values does not change if I change the configuration to differential mode. (INx- is grounded).

These observations does not match with the fact that the chip supports 2VRMS in differential mode and 1VRMS in single-ended mode (When using VREF=2.75V). But only the half.

Please explain.

BR

Kasper

  • Hi Kasper,

    I'm not sure why you are measuring significantly lower bias voltages than expecting unless the loading is significantly greater than 20kOhms or the 20kOhm load has actually been placed on VREF. Your AREG and DREG voltages are correct. I have measured these voltages on our EVM and they come out much closer to expected values.

    It sounds like there may be an issue in your configuration though because with the same input level you should measure a 6dB difference when switching between single-ended and differential input modes. 

    Are you using an EVM for these measurements?

    Best,

    Zak

  • Hi Zak

    I am not able to use the EVM. We got a brand new ADC3140EVM-PDK, but we have no software to control it. I have requested it by using the registration formula (filled in correctly and truthfully!!), almost a week ago. But no response at all. By the way, the device marking are different for the ADC on the EVM (XDC3140) and my prototype (ADC3140). Should I care about this?

    Regarding question 1:

    We have measured the Vref on two different prototypes and they are identically. One of prototypes generates following Vref and MicBias in different configurations. It seems that when MicBias is unhooked from Vref, Vref becomes slightly higher.

    BIAS_CFG    Vref  MicBias
    0x00      2.445V   2.317V
    0x01      2.225V   2.112V
    0x02      1.195V   1.164V
    0x10      2.445V   2.531V
    0x11      2.225V   2.308V
    0x12      1.195V   1.272V
    0x60      2.467V   3.290V
    0x61      2.240V   3.290V
    0x62      1.208V   3.290V
    I have a 30kohm (not 20 as written before) load on MicBias, and only the 2µ2F on Vref.

    Are there any other registers that can affect the Vref and MicBias.  (Besides the sleep register, of cause).

    Regarding question 2:

    I just found out that there per default are enabled 2 biquad registers per channel. In the time of writing I have not tried to disable them. But should it do a difference.

    Do you have a description of the digital format? (Besides the packing in I2S/TDM.) I believe that the data is presented in 2s-complement i differential mode. But it is not written anywhere in the datasheet. But have are the data formatted in single ended mode? And how are the full scale output range arranged in the data.

    Best regards

    Kasper

  • One more observation. I observed that the MicBias was lowered when I simultaneously measured Vref with a simple multimeter, therefore you must take into account that the Vref pin has a rather high impedance. The 2u2F I got on Vref has a bad insulation resistance. Around 20Mohm. I replaced it with one with a insulation resistance at 100Mohm. This change raised the Vref a little, but not significant.

  • Hi Kasper,

    Have you received approval for the EVM SW yet? If not please share the email associated with your myTI account and we can make sure you are approved. The difference in device markings is inconsequential. It is true the VREF pin is not meant to be loaded, but you should be able to measure this without significant droop. Are you sure you are not writing to reserved registers in the device? 

    The digital format is selectable but is always a 2's complement output with codes ranging from 80000000 to 7FFFFFFF and scaled such that a 1Vrms signal translates to 0dBFS.

    Best,

    Zak

  • Hi Zak,

    I did a new request last week, and I got the approval the day after. And it was on the PDK I discovered the high impedance on Vref.

    About the full scale, I therefore can conclude that when I use single ended input, the ADC only converts data to 00000000 to 7FFFFFFF. To summarize (which I really missed in the datasheet):

    At Vref=2.75V:

    Single-ended input  (Vin- grounded): 1Vrms gives converting results in the range 00000000 to 7FFFFFFF

    Differential input: 2Vrms gives converting results in the range 80000000 to 7FFFFFFF

    So basically the ADC does the same in single-ended input mode as in differential mode with Vin- grounded. As well as the DC or AC coupling only differs on the driving strength of the inputs towards Vref/2. (Also not very clear in the datasheet) And for that matter I just read in the descriptions that 2.5kohm input impedance is only supported in AC couple mode. This info is missing in the register descriptions. And by the way the tolerances on the Vref are neither specified in the datasheet.

    Regarding my initial questions I am not closer to an solution. Regarding the inaccurate Vref I have ordered some samples on the ADC. Just to be sure that it is not a faulty chip. And regarding the input range span, I will try to modify the PCB and "release" the Vin- inputs from ground and connect them to Vref/2 allowing me to use differential mode, and therefore the the signal will theoretically converts to values from C0000000 (i.e. -40000000) and 40000000. Maybe the the input range issue is no solved, but at least the upper part of the signals are not clipped, and we are not halted in the development.

    BR

    Kasper

  • I just made some more measurements. In differential mode maximum signal is around 2Vrms. But in single-ended mode it is only a quarter of it. 0.25Vrms. First of all the digital range is only the half, as no negative values are converted. But secondly The maximum DC-voltage that can be converted is only Vref/2, where in differential mode the span is -Vref to +Vref equals 2*Vref.

    My setup is following. I got 4 signals all with an amplitude at 0.77Vrms ~ 2.2Vpp but with different frequencies.

    Vref = 2.5V. Channel 1-3 has a Vin- at Vref/2 (voltage divider on MICBIAS), and channel 4 has Vin- at Gnd.

    When all channels are in differential mode, I have following output from the ADC: (Values multiplied by 2.5V/2^31)

     All perfect.

    But now I alter channel 4 to single ended input:

    So the conclusion must be that the full input range in single ended mode is Vref/2. Allowing a maximum signal amplitude at 1.375Vpp |~ 0.5Vrms at Vref=2.75V.

    The values on channel 4 in the latter chart is between 0x04899177 and 0x7FFFFFFE

    In other words it seems that the designers wanted to use the full digital data range in single-ended mode, but forgot that it was not possible when using 2s-complement.

    When changing to differential mode also solved out in accurate Vref. So this is our solution.