This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

LMP8358: How can I reduce the settling time with gains=100 or 200

Part Number: LMP8358
Other Parts Discussed in Thread: MSP430G2553, TLV2780

Hi,

I am trying to measure strain gauge with LMP8358 in serial mode. In my application, it is very important to reduce power consumption as much as possible. 

As a result, I shut down the LMP8358 and strain gauge most of the times and enable them periodically (every 20 ms for 50 samples/sec) to measure the strain gauge value.

When I want to measure strain gauge, at first, I enable LMP8358, wait about 100 us (according to data sheet,  typical required time is 85 us) and then I excite the strain gauge and wait for output settlement

and then read it. The strain gauge consumes lots of current and it is critical to reduce the IA settlement time to minimize excitation time and thus, power consumption.

According to data sheet, with gain=1000 and comp[2:0]=011b Ts=4 us which is great. Unfortunately, gain=1000 is to much for my application and reduces the strain gauge dynamic range significantly.

I want to use IA with either gain=100 or gain=200. I tried different compensation values with these two gains but the settling time is much higher than 4 us.I tried these settings:with gain=100 ,comp[2:0]=010 ,100

(suggested comps for this gain in data sheet) and with gain=200, comp[2:0]=010, 011, 100. In all cases the settling was around 15 us to 20 us.(much higher with 4 us).

Am I missing something or the chip has slower responses in lower gains?

  • Hello Amir,

     

    I am looking into this and asking around for more information on this part.

     

    What is the input signal's magnitude when you excite your strain gauge? What is the load you are driving with the LMP8358? How are you measuring the settling time exactly?

     

    The 4us settling time (ts) spec is under the conditions of a 2V step (so 2mV input to get 2V output at 1000x gain) and a capacitive load of 10 pF. Technically, any comparison on settling times for various gains should be tested under the same conditions listed in datasheet.

     

    If you’re inputs are being momentarily stretched apart by a few hundred millivolts then slewing at your output could occur, which would increase the settling time comparatively to the how the 4us ts specification was derived. Note no slewing occurred when inputting a 2mV signal to get a 2V output at a gain of a 1000x.

     

    The settling time for gains other than 1000 are not listed in the datasheet and spec table, so TI cannot say what these times should be. I would suspect, in general, that with lower gains (thus higher bandwidth), that the settling time of a system would decrease; however, I would need more information for further analysis.

     

    We can also try to search for a part with a faster settling time, if you have want to share a couple key features needed for this part. 

     

    Best,

    Peter Iliya

    Precision Amplifiers Applications

    Tucson, AZ

  • Hi Peter,

    Thank you for your response. 

    The input signal magnitude with gain=100 is around 27 mv. (The output amplitude is about 2.7 v).  

    The input strain gauge has balanced full bridge configuration  and each LMP8358 input voltage is very close to Vdd/2 (with respect to ground). Vdd=3.56 v.

    Bridge resistance is 350ohm.

    I use LMP8358 with msp-exp430g2 lunchpad (microcontroller P/N msp430g2553). The output of LMP8358 is connected to pin2 (P1.0) of micro which

    I configure as ADC10 input (channel A0) to read the value of strain gauge. According to MSP430x2xx Family User's Guide page 538 the input stage of

    ADC10 is equal to a 27 pF capacitor in series with a maximum of 2 Kohm resistor (MUX on resistance). 

    The supply is provided by launchpad and measured as Vdd=3.56 volt. I use a 100 nf decoupling capacitor directly between GND and Vdd pins of LMP8358.

    The capacitor is located as close to pins as possible.

    I just included the amplifier response to strain gauge signal together with the strain gauge excite pulse. CH 1 is amplifier response (500 mv/div) and CH 2 is the

    excite pulse (5 v/div). Amplifier Gain=100 .Comp[2:0]=010b. MUX[1:0]=00b. CUR[2:0]=000b. POL=FIL=PIN=0b.

    I would appreciate if you help me find a faster IA. As I mentioned, low power consumption is critical for my application. so I need an IA with

    shutdown feature. Since I monitor strain gauge every 20 ms, it is important that IA could wake up very quickly and has a fast step response

    to excite signal to reduce the excitation time as much as possible.Preferable gain for my application is between 50 to 200.

    Thanks, 

  • Amir,

    Just so I'm clear. Was this oscilloscope waveform taken as the ADC was sampling and reading the data? What is your ADC sample rate? Also, what are you doing to drive the reference pin of the device?
     
    I have some settling time data for other gains and it looks like it should settle within 4us, but I just want to be sure that the testing scenarios match up.

    Peter Iliya

  • Peter,
    Yes, this waveform is taken while amplifier output is connected to ADC input.

    The sampling time is 16*ADC10CLK=16*62.5 ns=1 us. (ADC10CLK frequency is 16 MHz). According to msp430x2xx user guide, it takes extra

    13*ADC10CLK to convert the sampled analog signal. So the total sampling + conversion time is (16+13)*62.5 ns=1.8125 us.

    In my application, after exciting the strain gauge, micro waits 12.5 us and then takes four consecutive samples from amplifier output.

    After that, it turns down excite for preserving the power.

    I even disconnected the ADC10 from output but I did not notice any visible change in waveform. Since, based on user guide page 538 , when

    the signal is not sampled by ADC10, its input is high impedance so it has no effect on amplifier output.

    Thanks,
    Amir
  • Amir,

    Thank you for all of the details you have been providing. These are helpful.

    How are you driving the reference voltage pins? Are you centering your output near Vdd/2 or at 0V? If your strain gauge is outputting a differential 27mV once excited with 5V, then at a gain of 100, your output will go to 2.7V. Thus according to your scope image, you are referencing REFF pin of the device at 0V.

    If this is the case, then you are violating the output conditions of the device in your single supply configuration since the output cannot be truely driven to ground potential.

    Peter
  • Peter,

    I put REFS and REFF pins voltage at Vdd/2 by connecting them to a buffered resistor voltage divider.

    So the amplified differential input signal can swing in any direction around Vdd/2 at output.

    As I mentioned previously, in my application, Vdd=3.6 volt and not 5 volt. 

    Amir

  • Amir,

     

    If the reference is being driven to mid-supply (1.8V), then the differential input signal you are measuring once the stain gauge is turned on is 9mV, not 27mV. According to your scope image (2.7V-1.8)/100 = 9mV = Vin at a gain of 100.

     

    Regardless after speaking with more team members, this is not your problem. Before you turn on your stain gauge your VCM is 0V and once you turn on the gauge with the quick 5V step (~10ns), the VCM of LMP8358 also quickly changes to 2.5V.

     

    There are two things to consider about this. One, this is not how the settling time tests were executed. In these tests the VCM does not quickly move, only the differential voltage across the IN- and IN+ that the part is sensing quickly moves, and the time the output settles is measured. Two, when you quickly change (close to 100MHz) the VCM of the part, you are stressing the common-mode rejection of the part (CMR). This means that the offset of the inputs will vary more as the VCM varies and this error is multiplied by the gain. This can very easily saturate your output and cause the wild oscillations seen. The CMRR for a 100MHz VCM variation is not even listed in the datasheet plot, but if you presume that at 100MHz the CMRR is 20dB, then your Vout = gain*CMRR(V/V)*VCM = 100*(10^(-20dB/20))*2.5V = 25V.

     

     

     

    The oscillations on your output should be the reason why the settling time is longer than typical, and if the oscillations are not due to the ADC load, then must be due to the input signal, since you are you are using internal compensation networks that have not shown to create this type of resposne. You can prove this by slowly ramping the 5V voltage of the stain gauge and observing if the oscillations disappear on the output.

     

    Peter Iliya

    Precision Amplifiers Applications

    Tucson, AZ

  • Peter,

    Thank you for your detailed explanations.

    you are right. The input differential signal must be 9 mv so the amplified input plus 1.8 volt offset set output at 2.7 v.

    Actually, I did some modifications in both firmware and hardware and get much better and faster response.

    First, I changed Comp[2:0] to 001b instead of 010b. this reduced the overshooting significantly.

    You can see the amp response after this modification.

    CH1 (500 mv/div) is amp output and CH2 (5 v/div) is excite pulse. Since, right now, the settling time is much shorter I could

    reduce the excite duration from 20 us to 10.5 us to save more power (Now, micro waits 4 us and then read four samples).

    I think the short spike at the beginning of amp response is due to CMR issue that you mentioned. 

    Another problem was the settled output value was changing continuously (peak to peak changes went up to more than 100 mv)

    from one excite pulse to another.(this is not visible in oscilloscope image)

    I noticed this problem mainly arise from the reference circuit since the reference was oscillating at around 1700 Hz and added up to output 

    variations. I added an RC filter (R=220, C=1 uF) between buffer output and REFF pin.This almost eliminated the oscillation on REF pins.

    However; still there is small variation at settled value of consecutive excite pulses (peak to peak of variation is about 40 mv)  that

    reduces measurement accuracy.

    I checked the REFF and REFS pins as well as Vdd and  inputs of amplifier wave forms with scope and everything looks fine and stable.

    It seems this is the amplifier noise itself. Do you have any idea about it and how I can fix it?

    Amir

  • Amir,

     

    Great to see that you have some better results.

     

    40mV p-p output noise does seem high. I would make sure you are measuring noise (whether for the output, REFF, REFS, or Vdd) correctly. You need to determine that the noise floor of your equipment is below the noise level of the circuit. Please look into our free course content on noise and how to measure it here: http://www.ti.com/precision-labs

     

    If you still need to find ways to reduce noise, you could try to lower the BW of LMP8358 by changing the compensation to 000 or 1xx. You could also add a low-pass RC filter on the output. This filter serves two purposes as mentioned in the datasheet here:

     

     

     

    Peter Iliya

    Precision Amplifier Applications

    Tucson, AZ

  • Peter,

    I do not think the problem is the equipment noise. Since, I am not seeing such variations or noise in waveform of other points.

    Also, I designed and prepared another instrument amplifier (IA) with op amp (P/N TLV2780) and resistors using

    conventional three op amp circuit for IA.

    The gain is 92.I connected this IA to strain gauge and it works fine. There is no noise or variations at the output of this IA and settling time is also 

    good (around 9 us). There is no added compensation capacitors or RC filter to this IA. (I may refine the response later by adding them).

    I prefer to use LMP8358 since I can set the gain and compensation by firmware and also CMRR and input offset voltage and offset temperature

    drift are better than that three op amp IA. 

    I added an RC filter to LMP8358 output. R=120 and C=10 nF. this creates a pole at Fc=1/(2*pi*120*10n)=133 KHz. But this did not solve

    the problem. I do not want to further reduce Fc or change Comp[2:0] since this slow down the response and increase the settling time.

    Still, I need to find some ways to reduce output variations.

    Amir

  • Hi Amir,

    If the variations you are seeing are truly from noise, then the only real option you have is to reduce the bandwidth of the system.  Something you may consider is using a non-linear filter, which essentially consists of a traditional RC-filter with antiparallel diodes across the resistor. I have attached an old Burr-Brown application report that walks through a few different topologies for implementing this. This allows you to achieve significantly better settling time than the conventional RC circuit for the same cutoff frequency. As Peter mentioned though, 40mV seems quite high to be coming purely from the LMP8358, so I suspect there is something else in your system producing this inaccuracy.  To clarify, you confirmed that this small signal oscillation is not originating from the reference voltage?

    AB-022.pdf

  • Hi Zak,

    Thank you very much for your interesting article about nonlinear RC filter.

    As I mentioned to Peter, I believe the variations I see at the output of LMP8358, come from amplifier itself and not reference. 

    Because, I checked the REF pins waveform and there is no oscillation on these pins. At first there were some oscillations at REF pins that by 

    adding an RC filter at the output of buffered voltage divider (input of REFF), were eliminated.

    Besides, the noisy variations at output start when LMP8358 goes to active mode and then , when the chip is shut down after measurement,

    the variations are also  disappeared and pure and clean reference value appears at the output.

    I tried to reduce the effect of this variation by taking several samples from output and then averaging them. However, this increases power consumption.

    Also right now, I am using wires to connect LMP8358 prototype board to msp430g2553 lunch pad which may contribute to noise. I hope by

    integrating all components on a well designed PCB and using battery instead of lunch pad power ,can further reduce the noise effect.

    Amir