"10bit ADC of LK24" vs "12bit ADC of LC60" ?

[BasePointer]

Hi,

New mcu of freescale, LC60 has 12 bit ADC. But the old one LK24 has 10 bit. Can you compare these ADCs for me? Is 12bit adc of LC60 really 12 bit?

What are the errors specified for LC60 below?

Unadjusted Error
Differential Non-Linearity
Integral Non-Linearity
Zero-Scale Error
Full-Scale Error
Quantization Error
Input Leakage Error

Monotonicity and No-Missing-Codes guaranteed in 10 bit and 8 bit modes

10x

[bigmac]

Hello BP,

The ADC of the LC60 is capable of providing 12-bit resolution, but this usually does not correspond with 12-bit "accuracy".  The various sources of error are explained in the data sheet, section 15.6.2.

To directly compare the error performance for the two devices, you should do the comparison with 10-bit mode for the LC60.  This way the LSB incremental size will be identical for both devices.  For 12-bit mode, the LSB incremental size will reduce by a factor of four, making any direct comparison more difficult.

Regards,

Mac

[BasePointer]

Hi,

Which of them has better accuracy?

10 bit LK24?

10 bit LC60?

12 bit LC60?

I know the datasheet defines the errors in my previous message. But I don't understand anything from there. I need a single parameter, that is accuracy. Can you compare them? Does 12 bit LC60 provide me more accuracy? I just wonder this.

10x,

BP.

[bigmac]

Hello BP,

The error components that you list are explained in some detail within AN2438.  The "absolute accuracy" for the LK24 should have similar meaning to "total unadjusted error" (TAE) for the LC60.

It would appear that 12-bit operation is not guaranteed for the 64-pin package version of the LC60.  To compare 10-bit and 12-bit operation for the 80-pin version -

10-bit mode:  Typical TAE = +/- 1 LSB.  These are 10-bit LSBs.

12-bit mode:  Typical TAE = +/- 3 LSB.  These are 12-bit LSBs.

To compare the two LSB sizes, one 10-bit LSB is equivalent to four 12-bit LSBs.

Regards,

Mac

[BasePointer]

Hi,

AN2438 is nice, thanks for that.

So we can say that:

LK24 10-bit mode:  Typical TAE = +/- 6 LSB.  These are 12-bit LSBs.

LC60 10-bit mode:  Typical TAE = +/- 4 LSB.  These are 12-bit LSBs.

LC60 12-bit mode:  Typical TAE = +/- 3 LSB.  These are 12-bit LSBs.

LC60 12-bit mode is better as much as 1 LSB (12-bit LSBs) than LC60 10-bit mode.

Also, LC60 10-bit mode is better as much as 2 LSB (12-bit LSBs) than LK24 10-bit mode.

Do I understand right?

10x,

BP.

[bigmac]

Hello,

The only additional comment I would make is that the error specification for the LK24 is a maximum value, not a typical value, as for the LC60.

Regards,

Mac

[BasePointer]

Hi Mac,

Finally, can we say that, 12 bit mode doesn't provide advantage much and approximately all modes are same?

Thanks,

BP.

[bigmac]

Hello,

It ain't necessarily so!  It depends on the specific application.

In many instances, the extra resolution that 12-bit mode provides is an advantage, where small changes need to be detected, and where the absolute calibration is less important.  Digitization of speech comes to mind - the dynamic range would be increased by 12dB compared with 10-bit mode.

For other applications, a calibration process can compensate for some of the measurement errors (such as zero scale error and full scale error), and 12-bit mode will then give more accurate results than 10-bit mode because of the increased resolution.

Regards,

Mac

[BasePointer]

Hi Mac,

We think of staring a new project to measure 3-phase active energy directly with internal ADC of LC60 such as this application: http://www.freescale.com/webapp/sps/site/prod_summary.jsp?code=RDS08LC60

We need to measure 30mA with error max 10 % and 100A with 1 % at all phases (via current transformers). I have some doubts if it is possible with the internal ADC.

I also wonder that, we can use 1.0 V or lower for VrefH and VrefL instead of MCU voltage? Can this cause any accuracy or linearity problems?

10x

BP.

[bigmac]

Hello BP,

---

BasePointer wrote:

I also wonder that, we can use 1.0 V or lower for VrefH and VrefL instead of MCU voltage? Can this cause any accuracy or linearity problems?

---

According to the data sheet you cannot do this.  VrefL must be tied to Vss, and the allowable range of VrefH is 1.8V  through to Vdd.

---

BasePointer wrote:

We think of staring a new project to measure 3-phase active energy directly with internal ADC of LC60 such as this application: http://www.freescale.com/webapp/sps/site/prod_summary.jsp?code=RDS08LC60

We need to measure 30mA with error max 10 % and 100A with 1 % at all phases (via current transformers). I have some doubts if it is possible with the internal ADC.

---

There will be insufficient resolution at low instantaneous current levels if you attempt to achieve current measurements in a single range for each phase.  However, it would seem feasible to achieve the required accuracy with both a high range and a low range for each phase, requiring a total of six ADC channels for the current measurements.  The analog signal conditioning requirements will consequently be relatively complex, involving multiple rail-to-rail operational amplifiers to provide buffering and level shifting.  The block diagram shown shown at your link seems to also imply this method (although I can't seem to display the magnified version of the diagram to see more clearly).

For 100A (rms) full scale input, this would correspond to a sinusoidal peak current level of 140A.  However, the waveform may depart from being sinusoidal.  For the purpose of discussion, I will assume a worst case peak/rms value of 2.  This would then imply that the high range ADC channels would require a peak-to-peak range of 400A, resulting in 12-bit resolution of 100mA, and giving significant quantization error with low instantaneous current levels.

If you were to then choose a peak-to-peak range of, say 4A, for the low range channels, the resolution would be 1mA, probably adequate for 30mA rms +/-10% measurements.

No doubt the reference design, when it is published, will cover these issues.

Regards,

Mac

[BasePointer]

Hi Mac,

As you said, the datasheet indicated min VrefH as 1.8V. This is problem for our current transformers, or we need an extra opamps that mean cost for each phase.

Is there any way to use 12 bit ADC at 16 bit or higher accuracy?

We also need to realize this in a single range.

10x

BP.

[bigmac]

Hello,

Since the ADC expects a unipolar input, and the CT output gives both positive and negative swing, signal conditioning will be essential, at minimum to provide a voltage offset to suit the operating range of the ADC channel.

To achieve the accuracy you require, your choice is to either use the dual range approach previously outlined, or to use an external 16-bit ADC device having three channels, and perhaps with SPI interface.  It is likely that some signal conditioning will still be required for the external device.

Regards,

Mac

[BasePointer]

Hi again,

The signal conditioning can be easly done without using an opamp by adding a dc offset. Also we can adjust the gain of CT(current transformer) with a load resistor. So in normally, we don't need an extra opamp. But if we need more gain, CT has limited load resistor. We can achieve this by chaning our CTs.

I hear that there are some methods to increase the accuracy of the 12 bit ADC to 16 bit accuracy. It called as oversampling. Is this method really sufficient, and can I use it for our requirements such as single range, 10% max error at 30mA and 1% max error at 100A? 

Does freescale have any application notes for it?

10x

BP.

[bigmac]

Hello,

Perhaps you should wait until you can examine the power meter reference design document.

Regards,

Mac