Faraday Technology design and manufacture filters for electronic signal processing. The design of filters is often referred to as a black art. With computer aided design it is often relatively easy to produce a filter circuit which satisfies the electrical specification. The main problem, and the reason why many organisations do not design and make their own filters if the realisation of the design in size, package type and price required, with the miniaturisation of electronics the demand for smaller and smaller filters has increased. Thirty years ago it may have been acceptable to manufacture a filter using silvered mica capacitors and large pot cores ending in with a box measuring 10 cm square. Today the demand is for the same performance in a SIL or surface mount device one tenth the size and able to pass through a reflow oven and be aqueously washed. The basic parameters which should be considered when defining or selecting a filter are,
Filters may be divided into five basic types and categorised according to the technology they use to realise the design that makes them work. The five basic types described in terms of frequency are
low pass, high pass, band pass, band stop and all pass. Although the descriptions are
self-explanatory the illustrations below describe each and define some of the basic terms
used to specify a filter. It would be convenient if the same physical technology could be used to make filters of any frequency but this is not possible so a range of methods have been developed over the years according to the frequency range. Analogue filters range from fully active e.g. using transistors, operational amplifiers and integrated circuits through passive components, inductors, capacitors and resistors to resonant cavity where the filter is a series of cavities milled out of a block of metal. A summary of the manufactory technologies and frequencies over which they may be used is given as a general guide only.
The range of frequencies covered by a particular technology is also dependent on the relative bandwidth. Until the 1950s filters were designed using what is called Image Parameter Theory. This theory is based on looking at the impedance of an electronic network to produce a prototype circuit. It is then possible to cascade many of the prototype networks together to increase the complexity of the filter. This method had many problems and in 1923 Zobal published a paper describing a new method of designing filters. This consisted of producing a polynomial equation to describe the ratio between the input voltage level and output voltage level in terms of frequency. This is referred to as a transfer function. By solving the polynomial and calculating the component or element values a much more exact design may be produced. This method is called Modern Network Theory. As modern network theory involves prodigious amounts of calculation to solve the equations it only became a practical method when computers were generally available. It would be ideal if a filter could be made to pass a certain frequency band but then immediately reject an adjacent band but this is not possible. There is always an area or band transition between the region of pass and the region of reject. The slower this transition is the simpler the filter can be. A number of standard transfer functions were derived by various researchers to solve specific problems. These have become standard topographies and most relate to the speed of the transition from the end of the passband to the start of the stop band. The most well known are shown below in terms of Loss and group Delay against Frequency. The fundamental difference is the cut off rate and the shape of the group delay response. These basic topographies may be used whatever the final method of realisation of the filter.
When considering a filter it is important to give consideration to the degradation the filter produces in the signal that it is required to pass. The effect may be categorised as follows,
The insertion loss may be corrected using amplifiers but the insertion delay is inherent and cannot be removed. The other distortions can be corrected by adding amplifiers and or equalisers. An additional consideration is the Time Domain Response. This is often missed because most filter design is done in the frequency domain, but as real world operates in the time domain when a filter is introduced into a circuit the results can be poor due to time domain ringing, even though the filter is working in the frequency domain as specified. All filters engineers have received panic calls from customers saying "but my system has ringing, what can you do about it?" The answer is often to use a different and softer roll-off filter. As a basic rule the faster the filter cuts off in the frequency domain the more time domain distortion and ringing it will introduce. Group delay equalisation moves the energy of the ringing to be more symmetrical about a pulse of impulse which make the effect more acceptable but it is a common fallacy to think that ringing can be removed by equalisation. With analogue filters it is important to specify a source and load impedance. In low frequency and audio systems it is normal to have high input and low output impedance. This is easily achieved when active elements are employed. At higher frequencies where inductors and capacitors are used it is normal to specify a transmission line impedance. With this type of filters it is important realise that power cannot be absorbed (except due to the inherent resistance in the components) and a filter rejects by reflecting the power back down the line. This is achieved by the filter changing its impedance at the interface with the source. A good source resistance is required to absorb the reflected power. |
Faraday
Technology Limited
Croft Road Industrial Estate
Newcastle-under-Lyme
Staffordshire
England
ST5 0QZ
Telephone
: (+44) (0) 1782 661501
Facsimile : (+44) (0) 1782 630101
E-mail : sales@faradaytech.co.uk