Problem
We needed sixteen high resolution differential analog inputs for a project we were designing. Signal levels would be in the low microvolt range so the board layout had to be as noise free as possible.
Solution
Our electronics design used very low noise instrumentation amplifiers and filters in the front end. This produced relatively high signal to noise ratios through the rest of the circuit. The front end circuit layout was on a single layer with minimum trace lengths and no vias. All circuits were surrounded by a ground plane and guard bands, with ground planes below on all layers. Following the amplifiers were signal conditioning circuits. The basic input circuit was repeated sixteen times, occupying half of the board area.
At the core of the design were four 24-bit Analog to Digital converters and their associated circuitry. These were positioned between the inputs and at the opposite edge of the board from the power supply circuits to achieve better noise isolation. This presented a problem for routing the analog signals from the more distant inputs close to the power supplies and digital circuitry. These input signals were routed on inner layer 3, sandwiched between local analog ground planes on the bottom layer 4 and inner layer 2. On inner layer 3 each trace was surrounded by a ground guard band to prevent signal crosstalk between inputs.
Each A/D circuit had its own analog and digital ground planes, connected via star routes to the respective power supplies. Digital signals from the A/Ds were routed between digital ground planes to four CPUs, each processing the output from one A/D. The digital circuitry had separate ground planes also star routed to the digital supplies. The CPUs connected to a RS-485 communications network to talk to the rest of the project's devices.
Power supplies were a problem. Most were linear supplies connected to the common +24VDC supply, and were inherently quiet. A -12VDC supply was produced by a switching inverter, a potential noise source. Further, this supply fed the input instrumentation amplifiers, so noise had to be minimized. We conducted extensive research and found what would normally be a very low noise part, with just millivolts of ripple. With filtering we tamed this supply so it produced less than 20 microvolts ripple at the output. Great care was used in routing the supplies and associated grounds to prevent noise from the digital circuitry from entering the analog circuits.
Results
The board was capable of achieving 20 real bits of raw data, which is the published limit for the A/D converters. With numerical processing the circuit could resolve signals as low as a few hundred nanovolts. This level of sensitivity was more than adequate for the project.
