Documentation : Hardware | Software | Tutorials    Wardrobe : Jackets | Applications
Papers : Notes | Proposals | Research    Administration : Stock | Minutes | Progress

Noise and Movement Models for Ultrasonic Receivers


This document outlines the discussion that Henk and Mike had last week with regards to improving the ultrasonics. The main point proposed is that if the received ultrasonic waveforms are examined in greater detail, we may be able to gain some information as to whether or not noise is present in the environment. This should allow us to give more information to mobile device movement models so that they may - more effectively - throw away erroneous readings.


The discussion was prompted by a few experiments that showed the waveforms of ultrasonic pulses observed at different stages of its propogation through the system. (Note that the following waveforms are crude representations of pulses observed on the oscilliscope and are meant for qualitative comparison only.)

The following waveform shows the ultrasonic pulse on the transmitter just before it is transmitted. The oscilliscope probes were attached to the leads of the transmitting speaker.

Transmitter Plot

The next waveform is the raw pulse observed at the leads of the receiving microphone.

Raw Received Pulse

This is the pulse after it has passed through the amplifying stage of the receiver (the input to the PIC).

Received Pulse after Amplification

And this is a plot of some noise generated by jingling keys at the same point in the receiver circuit (input to the PIC).

Noise after Amplification

Identifying Noise

The noisy plot is characterised by a large energy signature where the input to the PIC is active for a large portion of the displayed time-line. The clean pulse, on the other hand, displays a very tidy energy signature where the same input is mostly quiet apart from a distinctive rectangular (and finite) pulse. We believe that it may be possible to use the differences in the two signatures to improve our current (do we even have one?) mobile device movement model (MDMM).

Our current system does not differentiate between genuine pulses and noise in the environment - a major obstacle to its reliability. An initial solution to this problem may be to implement complex algorithms that extract the pulses from the noise. However, upon examination of the plots in the previous section, we conclude that this would be an arduous task (too big for a wearable application). Instead, we propose that the system need only determine whether or not there is noise present in the environment at the time it is taking its measurements. If there is noise, then the system can treat the readings accordingly - by throwing them out or weighting them with very little confidence. We believe that the signature of noise is differentiable enough from a "good" pulse to allow us to achieve this.

The Movement Model

Consider a MDMM that tracks the system's position through space and time. The model would be able to calculate its distance from each transmitter at any point. More importantly, the system would be able to predict when each of the individual pulses were expected to arrive. For a device with velocity and acceleration constraints, the system could construct an expectancy time window for each of the pulses.

Expectancy Window

Furthermore, the system would be able to predict the energy signature of each pulse since it knows the distance from each transmitter and the effects of attenuation (greater distances have pulses with smaller energy signatures).

The resulting model is one that is able to put greater confidence in its readings. Pulses found within the expectancy windows with expected energy signatures, for example, would be given high confidence levels while unexpected readings would be given lower confidence. We believe that a model that can do this will be able to effectively ignore reflections and noise.

Increasing the Frequency

If the predictive model is able to effectively eliminate reflections, it may be possible to increase the frequency of the system. Doing this will give our model a greater response time and accuracy.

It may also be possible to increase the frequency of the system based on the topology of the transmitter array. Transmitters that are far enough apart, for example, may be able transmit at the same time, removing the need for large transmission delay cycles for each transmitter.

Points / Questions

This page last updated November 13, 2002
The material displayed is provided 'as is'. Contents may not be reused without prior permission.
For problems or questions regarding this web contact Cliff Randell.