19-07-2012, 03:21 PM
Wireless Sensor Network to Determine Sound Origin
Wireless Sensor Network .doc (Size: 159 KB / Downloads: 33)
We chose to do this project because it is highly applicable in many real-world situations. This project will allow us to apply our knowledge of DSP, communications, circuit design, analog signals, and computer science. We're especially enthusiastic about seeing our theoretical training in practical use. We feel that the demonstration of our project promises to be interactive and fun.
Our goal is to implement a wireless network of microphones that will be able to make accurate measurements of the differences in time that it takes a sound pulse to reach various sensors within a room. The information will then be communicated to a PC, where calculations will be made to determine the sound's origin. Event locations will be flagged in an OpenGL model of the room, and may be indicated in real space with a servo-controlled laser pointer.
Applications of our project can provide several benefits to the customer:
• Find a member of a theater audience using a “clapping” cue with spotlights (rather than a laser).
• Improved security to locate intruders via a system independent of traditional optical sensors.
Extentions of our project are also be applicable to consumer problems:
• If the system matches up voice samples rather than impulses, a spotlight could follow a speaker around a room.
Product features:
• Locate the source of a noise without the need for hardware mounted on the source
• Easily determine the origin of a noise without visual cues
• Wireless communication allows for easy installation of the network.
• Simple reconfiguration of the network (i.e. Dynamic addition and removal of sensors during operation)
• Servo-controlled lasers will indicate the 2-D location of the source.
• Avoid the need for absolute timing in the sensor network
Design of the Sensor Network
The network will have a base and n sensor units positioned uniformly in the room (Figure 1). The lines indicate that the sensor units communicate with the base as well as with each other. The communication with the other sensors is so that the sensor units know when to start measuring the time delay in receiving the pulse. The sensor units then inform the base what each particular delay is. The base must communicate with the sensors so that they know to be in the “wait” state (for an impulsive noise).
Each sensor unit (Figure 2) will be composed of a microphone, whose analog output is immediately low-pass filtered. The signal will then be appropriately amplified so that signal processing techniques can be applied through a PIC microcontroller. The microcontroller and RF unit work together to send information to the other units in the array and to the base.
Like the sensor unit, the base unit (Figure 3) also has an RF and PIC working together. The information that is received and processed will be sent to a PC so that the calculations can take place. The final product is a set of 2-D coordinates that will pinpoint the sound's origin. That can then be sent back to the RF in order for the servo-controlled spotlight or laser to function appropriately.
Testing procedures:
We will test the accuracy of the microphone network by creating a sound at various known locations. We can then determine exactly how far off the calculated position is from the actual position in the room.We will also test the sensitivity of a sensor unit to the sound of a clap or snap. We will do this by creating a sound from various distances and finding where the sound is only detected perhaps half of the time. In order to be consistent, we may choose to do this testing by playing a sound recording.
Tolerance Analysis:
The main feature of this project that can affect the performance of the system is the number of sensor units placed throughout the room. We would like for the system architecture to support many sensors, but we also would like to have respectable accuracy when there is less timing information to make calculations with. In order to gain better accuracy, the network will rely on redundant information from the sensors. It will then attempt to calculate a most probable location based on the timing information given. With fewer sensor units, there will be less timing information available and this will restrict the ability to narrow down the region where the sound event may have occurred.
I think we can expect calculated positions to have a Gaussian-like distribution centered around the actual location of the sound source. The more confined this distribution is, the better the accuracy of the system. We will attempt to determine how many sensors are required to give, for example, 90% probability that the event actually occurred within 1 foot of the calculated position.