0

I'm designing a signal acquisition system with an ADC. I have these questions:

1- Any cons or pros related to connecting the analogue and digital planes using ferrites or inductors. Or should I just connect them directly with a trace/wire at one point near the main supply ground?

2- There is a document on Analog website that states:

"Be sure to properly decouple each supply pin at the device with a 0.1-µF ceramic capacitor. Remember to decouple to the analog ground plane, not the digital ground plane"

I find it odd that the digital supply decoupling cap is connected to analogue ground? Wouldn't digital ground be a better choice?

3- What ground do I connect the sensor shield wire to? I'm guessing digital because I wouldn't want the noise it brings in on my analogue signal. Is this correct? Edit: Or a third option, chassis ground?

Thanks

hadez
  • 579
  • 6
  • 14
  • 2
    Relevant info here: http://electronics.stackexchange.com/questions/34974/how-do-you-choose-an-inductor-for-connecting-separate-ground-planes – Shamtam May 08 '16 at 22:55
  • 1
    Re: #2... you should read that entire document. They specifically answer your question, see: "Q. So the right answer is to connect both AGND and DGND pins to the analog ground plane? But doesn’t this inject digital noise on my nice quiet analog ground plane?" – Shamtam May 08 '16 at 22:56
  • 2
    Much of the time there is no need to have separate planes. Please consider if you really need to have a separate analog and digital ground. – user57037 May 08 '16 at 23:01
  • It would help to have some background on why you're separating grounds in the first place. The Howard Johnson Black Magic book is a good place to start. – Daniel May 09 '16 at 00:06
  • @Daniel Actually I kinda got the answers I needed for #1 and #2 but question #3 still needs to be resolved. – hadez May 09 '16 at 00:27
  • @mkeith To separate the return paths of the digital signals and the analog ones from each other. I'm using a 24 bit precision ADC and need all the resolution I can get. I am not using two separate layers for each GND, just separate sections on the PCB. – hadez May 09 '16 at 00:32
  • I remain skeptical. To separate return paths implies that you are forcing return paths to go around an obstacle, which implies you are adding noise, and increasing radiated emissions. The most important thing is physical separation between victim and aggressor signals. Separating ground planes is usually a bad idea. – user57037 May 09 '16 at 00:37
  • @mkeith I am not going around obstacles. Imagine a few rectangular sections of copper on the PCB separated from each other using copper-less voids, and connected using star topology at one point. Each rectangle serves the ground for a separate part of the circuit. Is this good enough? – hadez May 09 '16 at 00:46
  • 1
    I don't know. Look at @Shamtam's first comment. In one of the answers to the linked question, there is a Henry Ott article that you should probably read. – user57037 May 09 '16 at 01:05
  • 1
    Look at it this way. If you have a split plane, why do you have a split plane? Are you trying to prevent current from flowing across the split? If current wants to flow across the split, that means there is a voltage across the split. If there is a voltage across the split, then current will be forced to flow around the split (creating a current loop with large area). That is bad. So you need to lay out your components so that no current wants to flow across the split. If you do that, then I think you can see that the split is not actually needed in the first place. – user57037 May 09 '16 at 01:10
  • 1
    Ground planes can be the subject of holy wars, but my take is at http://electronics.stackexchange.com/questions/185306/analog-power-ground-planes-doubts/185320#185320 – Peter Smith May 09 '16 at 08:46

1 Answers1

2

I have designed and built boards with mixed analog and digital, and they do have separate ground planes. (I used a 4 layer board) The ADC converter had separate analog and digital ground pins. (it was an LTC1605IN)

This is the point were Agnd and Dgnd were joined with a very short jumper wire. ALL analog sources were analog grounded at their inputs, including any shields for shielded pairs, as I often used 600 ohm STP (shielded twisted pair) to keep noise pick-up and common mode DC offsets to a minimum.

The Analog power supply used a small pc mount DC-DC converter to provide isolated +/- 15 Volts. In some cases I just use LM7805 and LM7815 and LM7915 linear regulators, so the power sources had the same source ground(at the power supply), but did not junction on the board except at the ADC converter.

It keeps the separate ground planes apart until they HAVE to join. The best place is the ADC converter, to prevent up-stream op-amps from picking up digital noise on the analog gnd trace. After the analog signal become digitized, they are standard logic '1' or '0'.

The digital logic IC's and CPU / MPU are much less sensitive to noise because an input has to cross certain thresholds at least 500mV to 4 volts apart before it is considered a logical '1' or '0'.