Access Official Technical PDF

Share Summary:LinkedInEmail

As sensor counts grow, wiring every sensing element back to a central controller can complicate installation and signal management. A distributed approach—digitizing readings close to the sensors and sharing them over a serial bus—can reduce the number of connections and help limit interference along long analog signal paths. A 48-node platform built for integrated MEMS sensors illustrates the design choices involved in reliable sensor network communication, from bus access rules to PCB transmission-line characteristics.

Why digitize at the sensor?

In a conventional scanning architecture, a central readout circuit handles sensor signals. As an array expands, the wiring and signal paths also grow, while crosstalk and scan time can become concerns. With local digitization, each sensor platform converts its inputs into digital data and sends packets over a shared bus. This can simplify wiring and reduce the distance that sensitive analog signals must travel.

The evaluated platform used a dedicated CMOS LSI that could operate in capacitive or resistive sensing mode. In capacitive mode, each LSI measured eight channels in parallel. Forty-eight devices therefore provided 384 capacitance channels on a 2.4 m bus. An FPGA-based relay checked received packets, including their CRC, and forwarded data to a host computer. The relay also passed configuration commands from the host to individual devices or to all devices.

Managing access to a shared serial bus

A shared bus needs a way to prevent multiple devices from transmitting at once. The platform used carrier-sense multiple access with collision avoidance (CSMA/CA). Before sending a packet, an LSI checked whether the bus appeared idle. While transmitting, it continued monitoring the line; if the observed logic level differed from the level it was sending, it identified a collision and stopped transmission.

This approach allows devices to share the line without a central schedule for every transmission. However, simultaneous attempts can still occur: two devices may both detect an idle bus and begin sending. The collision response limits the duration of a conflicting transmission, while an adaptation function helps manage repeated attempts and distribute bus access.

How the adaptation function works

The adaptation function applies two configurable waits. First, a device can delay a packet after sensing data crosses a configured threshold. As more packets enter this stage, the delay increases according to a set increment. Then, if the bus is busy, the device waits before trying again; this bus-wait interval also increases with each attempt. If the configured attempt limit is reached, the pending packet is discarded and the device moves on to new data.

An event-driven option can further reduce traffic by sending data when a channel’s reading exceeds its threshold, rather than having every device report continuously. In the reported 48-device evaluation, the adaptation function was enabled, but the event-driven thresholds were set to zero. The results therefore characterize simultaneous activity with adaptation, not a test of event-driven traffic reduction.

Signal integrity matters in a long bus

Bus-access logic alone does not guarantee reliable communication. The team built the system from three PCB units, each carrying 16 LSIs on a backbone bus with short stubs. The 2.4 m line introduced signal reflections, and the designers found that the bus characteristic impedance needed to be high enough to prevent those reflections from making an LSI falsely detect a collision.

Termination involved a trade-off. A 110 Ω resistor, close to the bus impedance, reduced signal fluctuation from reflections. But its lower resistance increased bus-driving current, which could cause greater variation in the 3.3 V supply when many devices were connected. A 10 kΩ termination reduced that current and was selected for the simultaneous-transmission evaluation. Under the tested 1 MHz communication speed, the resulting signal vibration did not prevent the devices from communicating.

The interface used single-ended signaling, which keeps the signaling arrangement simple but is less resistant to electrical noise than differential signaling. The successful test should be read in its specific context: the selected PCB layout, bus length, termination, device count and transmission speed. These factors need to be considered together when applying a similar architecture.

What the 48-node evaluation showed

All 48 LSIs were configured to operate simultaneously, creating a demanding bus-traffic condition. During the evaluation, the number of correctly received packets from each device remained broadly similar. Each LSI delivered between 1,890 and 2,315 correct packets in a 30-second recording interval, and the difference across experiments for an individual LSI was below 12.76%. The reported packet error rates were below 0.42%.

Across 384 capacitance channels, the measured average sampling frequency was 73.66 Hz. The median interval between successive packets from a given LSI was 9.76 ms; the mean was 13.58 ms. Most measured intervals were below 30 ms, though the maximum interval corresponded to a worst-case sampling frequency of 5.3 Hz. These figures distinguish average system behavior from the less frequent delays that can matter in time-sensitive applications.

Practical implications for system designers

For engineers planning a distributed sensor installation, the evaluation highlights several connected design tasks:

  • Choose the readout architecture: Local digitization can reduce wiring and long analog paths, but shared-bus capacity and packet timing still need to suit the application.
  • Plan bus access: Collision detection and configurable retry delays can help manage simultaneous transmissions, while event-driven reporting may reduce unnecessary packets when application thresholds are appropriate.
  • Design the physical layer with the load in mind: Bus impedance, reflections, termination, driver current and supply stability interact. Validate the complete line rather than treating termination as an isolated choice.
  • Evaluate worst-case timing: Average sampling rate does not describe every packet interval. Consider the maximum delay as well as average performance when assessing control or monitoring requirements.

The 48-node result shows that a backbone serial bus can support a sizable distributed sensor platform when communication rules and electrical design are developed together. Its practical lesson is not that one termination or access scheme fits every installation, but that topology, signaling, power stability and traffic behavior must be evaluated as a system. For applications where timing is critical, the observed worst-case delays and the stated limits of the collision-avoidance approach should be part of the design decision.

Access Official Technical PDF

About The Author