“I can’t quite decide”

…or low duty-cycle p/n dithered burst transmission in low-power radio networks! Numerous short-range data transfer applications now involve a basic architecture that’s considerably more interesting than a basic one-to-one pairing.In some cases the radio network is as functionally complex asa wired-Ethernet installation, with multiple nodes exchanging large amounts of data on a peer-to-peer basis. In others it can be as simple as a single master controller, periodically broadcasting a handful of command bytes to a constellation of subordinate receivers, with each controlling a physical device.The system type I intend to examine here is of a common variety, best identified as a “data gathering” network. In this case the system comprises a large number of sensor nodes,which could be sensing anything from complex environmental data to the state of an alarm button, all reporting to a central master station, where the data is gathered and processed.While there are many possible uses for this network type,such as fire or intruder alarms, agricultural monitoring,perimeter intruder detection and energy efficiency monitoring,they are characterized by a handful of common characteristics:

  1. Data flow is one way: from sensor nodes to master.
  2. Multiple transmitters share the same channel.
  3. Individual transmit duty-cycle is low, to minimise power
    and meet regulations.
  4. Overall data throughput is low; each node sends only a few
    dozen bytes.
  5. Response time is not critical; a ten second delay would not
    be disastrous.

Experienced users of low-power wireless devices will already be familiar with a number of techniques that address one or more of these requirements, such as traditional sequential polling, or beacon synchronisation, not to mention any number of more sophisticated proprietary mesh-network techniques.Unfortunately, all these methodologies require the use of a wireless device capable of functioning as a transceiver. While such hardware (in this age of single-chip radio devices) is no longer prohibitively large or expensive, a transceiver is still more expensive than the equivalent transmitter, and all these receiver function that maintains network synchronisation from the periodic base transmitted timing-burst.

How Else Can This Sub-Class Of Network Be Organised?

Obviously, if all transmitters are allowed to send continuously, as well as suffering from a prohibitive level of power consumption, they would all block each other and,allowing for capture effect, only the nearest node to the master would have a chance of being heard.Time-division multiplexing is the obvious solution. Organisethe data into a concise burst or packet, with the necessary framing, addressing and check-sum sequences,and then transmit it a sinfrequently as the overall system constraints allow;which could be anywhere from less than once per day, to several times per second. This technique obviously allows very low duty-cycles and, with some care in design of any periodic wake-up “heartbeat timers”, low-tonegligibleaverage current, all achieved with a very simple transmit-only node. Perfect? Not quite. All “transmit blind” methods retain the risk of any given transmission colliding (occurring in the same timeslot)with another, resulting in the loss of at least one set of data. This can happen when external events simultaneously trigger two or more nodes at the same time, or when by pure bad luck the periodic transmission “beat” of one node falls into step of another. In these circumstances the only solution is for each node to re-transmit the data enough times that at least one of its messages will reach the master receiver uncorrupted. For such re-transmissions to avoid simply becoming repeats of an initial collision, it is vitally necessary that no two or more nodes transmit with the same periodicity and pattern. A pseudo-random number sequence-generator is the key to a very effective method of achieving this: If, instead of sending a single data burst, the node sends a number of identical bursts, each separated from the previous by a (pseudo) random time period then the likelihood of at least one getting through is greatly increased. The chance of losing an entire data-set to collisions with another transmitting node becomes insignificant if making the number of repeats large enough.

How Many Repeat Transmissions Are Necessary?

 This is the critical question. The chance of a collision is related to length of burst, number of nodes transmitting on the system and the frequency of transmission. While direct analysis of the probability is possible, for practical engineering purposes I use a simple software simulation of the network. The required coding is very simple and can be implemented in any high-level language. The technique I suggest is the following:

Determine the length of the transmit burst. This is the  basic “granularity” or “time slot” of the system timing  (there is no point in a time step of 1ms if the bursts are  50ms long; a timing change of less than fifty “ticks”  would not avoid collision).    

  • Write a good software model of the p/n code that your  actual system is using (length of shift register, position  of taps, initial seeding, transmission rules etc). 
  • Set up a separate routine for each node on the system,  determining if it transmits in a given time-slot  (according to the rules set up above).
  • In each time slot simulated, record which nodes  transmit and so determine how often each node  successfully sends a burst without collision. 
  • Run the simulation over many thousand simulated  time-slots, to yield a statistically meaningful result. I  have used simple interpreted BASIC as my simulation  tool 

Once a valid simulation has been written, it is then  possible to vary the number of nodes, the frequency  of transmission and the number of repeats, until an  acceptable compromise is found.   

View PDF