Vref电路_大小电容并联

The aim is minimum noise on the reference channels VRH and VRL.

Sources of noise

Noise gets onto the reference supply from both directions, the incoming voltage and the disturbance from the ADC current:

Z1VCC5VRHVOFF = VAMPL = FREQ = 5VdcV2IvrhV1Z2orcad_ivrh.txtVRL0Generic filter

Noise from the left can be grouped into two types – r.f. and sub MHz frequencies. R.f noise comes from the digital i.c.s, memory bus etc. but also in the case of the ADC from the VDDA supplies. This is because the 5V source for VDDA is probably the same supply used for VRH, so the high frequency current spikes from VDDA will couple onto the references.

Low frequency interference will come from the ECU, from PWM drivers, inductive loads, communications links and the like.

Noise from the right is due to transients of current drawn by the ADC reference supplies.

Low frequency filtering

The higher the impedance Z1, the more effective the filter. But there is a limit to the d.c. impedance because there is a d.c. current drawn from VRH by the ADC. Each additional 5 ohms causes about 1 LSB of voltage drop, so the resistance of Z1 is limited to a maximum of around 10R to maintain the d.c. accuracy of the reference supply.

The lower the impedance Z2, the more effective the filter. With Z1 at 10R, we get the filtering below with capacitance of 10n, 100n, 1u and 10u for Z2.

10RVCC5VRHVOFF = VAMPL = FREQ = 5VdcV2IvrhV110n-10uorcad_ivrh.txtVRL0

It can be seen that only 10u and 1u provide any appreciable filtering of kHz frequency noise. Note that although the capacitor is performing supply decoupling, high currents are not involved which permits the use of lower cost capacitor types.

High frequency filtering

At the frequencies of the MCU, the parasitic effects of the capacitor are important. For example, Murata 1uF part GRM21BR71C105KA01 looks like the following impedances at 100MHz:

ESL440pESR24mC1u

Although a theoretical capacitor has an impedance that lowers with increasing frequency, above its self resonant frequency a real capacitor begins to increase in impedance again. This response is seen in the low frequency filter proposed above when the parasitics are added:

10RVCC5VRHVOFF = VAMPL = FREQ = 5VdcV2L3-esl300p-630pIvrhV1R3-esr3m-235mC310n-10uVRLorcad_ivrh.txt0

Filter response to 1GHz for 1n to 10u in decade steps, including appropriate parasitics

The point to note is that above 100MHz there is not much filtering! Even the 1n in 0201 package still provides only limited attenuation, insufficient to attenuate the noise from VDDA down to a level on VRH where it will not impact accuracy. Since Z2 cannot be reduced, Z1 must be increased. This can be achieved using inductance. Ferrite beads such as Panasonic EXC-ML32A680H (1206 package) are effective, and chip inductors such as Murata LQW15ANR10J00 (0402 package) remain inductive up to 2.4GHz. With these the filter continues to be effective at high frequency. The chip inductor with parasitics looks like this:

C5-para0.04p780m100nR5-paraL5-filterso the filter becomes:

C5-para0.04p10RVCC5780m100nR5-paraL5-filterVOFF = VAMPL = FREQ = 5VdcV2L3-esl300p-630pIvrhV1R3-esr3m-235mC310n-10uVRLorcad_ivrh.txtVRH0

And the result is:

Note that the 10R resistor must still be used to maintain low frequency filtering, since the inductor has no significant impedance at low frequency.

Reference supply decoupling

On the face of it, this arrangement 10R/100n/10u would appear to be sufficient. However the simulation model is not complete. Although the capacitor parasitic inductance is considered, there is also track inductance from the capacitor to the ADC. Whilst this does not affect the ability to filter incoming noise, it does impact the effectiveness of the decoupling of the ADC power supply.

Current drawn from VRH during one conversion.

The current spikes of the ADC are very fast, from 1-300MHz, so a low impedance decoupling capacitor is required to minimise voltage variation. Only 10nF is required to supply the current pulses associated with each bit of the conversion process so the existing filter has enough capacity, but the fast rise times will only be decoupled if the capacitor parasitics can be kept low. This is why a second capacitor is advantageous. It can be seen from the previous graphs that the 1nF capacitor has the highest

frequency response. This capacitor is also physically much smaller than the 10uF, and so can be placed closer to the ADC supply pins. This close placement minimises the track inductance and so maximises the effectiveness of the supply decoupling.

So the recommended VRH filter is as below:

10RVCC5100nL5-filterVRH5VdcV1C310uC41nIvrhorcad_ivrh.txtVRL0

VDDA filtering

The current drawn from VDDA is similar in pattern to VRH but higher in amplitude.

Current drawn from VDDA by one ADC over one conversion.

The current transients are fast so ESL is again the critical parameter for decoupling. Hence a close coupled 1nF is recommended.

A 1R resistor causes about 5mV of drop. Voltage drop on VDDA will cause a 5V ADC signal voltages to exceed VDDA, but the ADC will continue to convert accurately with small differentials.

10n is again sufficient to decouple the bulk current, with 100n and larger less than 2mV different. VDDA is less sensitive to noise than VRH, and does not require as much filtering. This is fortunate because with the lower resistance, less filtering is possible! As with VRH, since the low frequency filter is removing noise that is depends upon the application, Freescale cannot specify this filter. However, comparison with the 10R filter on VRH shows that 10uF will be required for any significant low frequency filtering.

Although the ADC amplifier has the ability to reject low frequency noise (power supply rejection ratio, PSSR), this only works within the frequency range of its

amplifier. High frequency will affect the accuracy – around 100MHz and upwards. For this reason an inductor will improve accuracy if the 5V supply contains high frequency noise.

Combined low cost supply filtering – MPC500 level performance

(L1)VCC510C31u5VdcV1R1VDDA1C41nR2VRHVRLC11uC21nVSSA

L1 must be at least 10nH, but this could be formed on the circuit board and would still filter the spikes from VDDA. However, a long track is also an aerial and will pick up r.f. noise if present inside the MCU. Very careful design would be required to avoid increasing the noise level on VRH. Without L1, spikes from VDDA will cause about 3mV of noise on VRH.

0Combined intermediate supply filtering

R2VCC510100nC310u5VdcV1R1VDDA1C110uC21nVSSAC41nL1VRHVRL0Spikes from VDDA are around 1/10 LSB on VRH. VDDA has little high frequency filtering.

th

Combined optimum supply filtering

R2VCC510100nC310u5VdcV1R11L2VDDA100nC110uC21nVSSAC41nL1VRHVRL0