Discussion of Filter Techniques
RS Microwave reserves the right to employ any technique known to the filter world. Our only products are filters. We use any and all approaches required to supply filters compliant to needs generated by our favorite people, our customers.
Some of our more common techniques are discussed here.
Evanescent (GR."Decaying Spirit") mode filters utilize the scattering of waves from obstacles placed in below-cutoff waveguide to form high Q inductively coupled bandpass filters. A waveguide structure has a normally high-pass frequency characteristic. By this is meant that frequencies lower than the cut-off frequency cannot normally proceed from one end of the waveguide to the other. The waves encounter a high impedance proportional to the dimensions of the waveguide. Note that any empty tube is a waveguide structure with cut-off characteristics dependent upon the cross-sectional dimensions of the tube. The tube may be of any geometric closed shape, including rectangular, circular, etc.
We can represent the high-impedance properties by a simple circuit. This "Pi" of inductors contains three elements. The shunt inductors are proportional to the cut-off frequency of the tube, while the series element is proportional to the length of the tube. If we resonate the shunt elements with capacitors and then cascade a number of these sections, we obtain the evanescent bandpass filter structure as shown in the figure below.
Well, so what? How does this help us to achieve superior filters? Let us look further at the properties of a rectangular waveguide. The dominant mode (lowest possible frequency of operation in the waveguide) propagates through the waveguide in a very low loss manner. We can say that the losses associated with resonance (necessary for building any filters) are very low. As we decrease the frequency below cut-off, the losses increase, but they were low to start with. Thus, although the evanescent filter operates below cut-off the associated losses are low. We express this by saying the unloaded Q (quality) of the structure is high.
Further, the waveguide will not allow non-resonated frequencies to propagate through until a frequency above the cut-off is reached. By building the evanescent filter in a waveguide small enough (i.e. operating far below cut-off), one can build low loss filters with no spurious passbands until some frequency above the cut-off of the small waveguide. It is possible to build filters with very high stopband to passband width ratios. In short, the evane-scent filters can have very wide stopbands and are thus suited to wide spectrum applications.
Such filters may be built with waveguide, connectors, pins or with combinations. Many such filters handle high power or are "dropped in" to receiver networks or are included in multiplexers, switched filters, etc. The technique is applicable from 10 MHz to at least 40 GHz, for bandwidths from 1% to 80%.
Lumped Element and Printed Circuit
Conventional lumped element bandpass filters are typified by the "dumbell" structures used in tubular bandpass filters or by the capacitively coupled approach commonly found in LC ("box") designs. Because these older structures are simplifications of an exact circuit, the responses are not symmetrical. This causes the designer to include extra elements to steepen either the lower or upper stopband slope. The extra elements take up size and add insertion loss. It is possible to synthesize filters in which inductor and capacitor elements alternate in the path connecting input to output. Through this means, the upper and lower slopes are made equal. The approach can be used for bandwidths to at least an octave, achieving lower loss and smaller packages as described above.
The LC approach is useful for not only bandpass but also for bandstop, highpass and lowpass circuits. Applicable from a few MHz to the mid GHz range, LC designs are flexible. Because the LC elements have Q factors in the 100 to 250 range, resultant filters are not suited for low loss or narrowband applications but find great application where gain is available or noise figure is already determined. At lower frequencies, the inductive elements are ferrite loaded or wound, and the capacitors are of the discrete chip or tunable variety. At higher frequencies (above a few hundred MHz) the inductors and capacitors can be printed, with either single or multilayer boards being used. Such multilayer circuitry usually requires the use of plated through "via" hole technology.
As frequencies increase, it becomes possible to combine distributed elements (quarter wave length resonators, short and open circuited lines, coupled lines, etc.) with printed L and C elements to achieve more optimum filter responses. For example, it is sometimes desirable to include additional stopband depth at frequencies near the passband. Traditionally, "elliptic functions" networks have been used in which the elements of the low-pass or high-pass prototype (the network which models the desired response) are resonated before being converted into a bandpass or bandstop structure. Such filters are difficult to build because of the wide range of element values involved. It is hard to locate a very small capacitor near a large inductor, for example. More recently, an approach has evolved in which the passband displays an equally-rippled (Chebychev) characteristic, but the stopband includes 3 transmission zeroes (frequencies of high attenuation) at an "infinite" frequency and an even number of transmission zeroes at frequencies near the band edge.
Such "generalized" Chebychev filters are almost as selective as elliptic function filters but are much easier to build, either as LC designs or as printed, suspended-substrate, networks. Combinations of such low-pass and high-pass filters can be used to make very efficient wide-band LC or printed filters, with 40 to 50 dB attenuation within 10% of a passband edge. In the printed designs, some of the elements are essentially lumped while others are intentionally realized using distributed elements. Thus, a design philosophy in which the natural responses of lumped and/or distributed elements are computer-combined results in better LC or printed designs. The printed approach is useful to at least 18 GHz.
Multiplexers are combinations of filters connected in such a way as to provide access to the passband and stopband characteristics of each filter from a common connection. Such a device permits the use of a single antenna with several receivers or transmitters, for example. The common port must display a low VSWR and isolation must be maintained between each of the component filters. A two channel version is called a diplexer, a three channel- triplexer, etc. (Fig. M-1). If the adjacent passbands of each channel "cross-over" at a level of about -3db the device is called a "contiguous" multiplexer. The frequency separation of channels is called the "guardband".
A multiplexer is normally used if a wide spectrum must be accessed equally and instantaneously. Conventionally, multiplexers have had the disadvantage of requiring at least 3 dB excess loss (”crossover” loss) at frequencies common to two channels. Thus, the passband characteristics for contiguous structures always showed an insertion loss variation over the passband of at least 3 dB.
To construct any multiplexer, it is necessary to connect networks to the constituent filters such that each filter appears as an open circuit to each other filter. While this is simple for narrow band channels, it is difficult for broadband or contiguous filters. Normally, the filters and the multiplexing network are synthesized as a set, with computer optimization being used to simulate the results before construction begins. Some of the more common multiplexing techniques include line lengths, circulators, hybrids, and transformers. More recently, the multiplexer filter channels have been combined using power dividers (Fig. M-2). This recent adaptation of always-available technology is due to newly-available cheap and compact amplifier stages. Such gain blocks provide flat gain and low noise over wide bandwidths.
In the case of two-way combining, conservation of energy means that the 3 dB insertion loss is still experienced...but on a flat-loss basis. Although each channel is subject to the additional 3 dB loss, it is essentially constant loss over each channel and thus the excess passband loss variation is less than 1 dB. Excess loss is defined as that loss not attributable to the individual channel filter roll-off. This power divider based combining can be extended to triplexers (4.7 dB flat loss), quadruplexers (6 dB flat loss), etc. Because the loss variation is minimized, the overall insertion loss can frequently be made up using amplifiers, which display flat gain versus frequency.
Filters can be multiplexed by parallel combination at both ends. For example, if two bandpass filters are multiplexed at both input and output, a network results which provides one input and one output, with two passbands, essentially attenuating everything else. Such assemblies are useful in systems such as GPS which have two or more operating frequencies, with the requirement for isolation between the operating channels and adjacent, cluttered regions of the spectrum (Fig. M-3). Another approach employs switched selection of filters (see section on switched filters). Hybrid combinations using multiplexers with power dividers, switches and amplifiers are now possible. The interactions of these essentially reactive components can cause undesirable degradation of stopbands or passbands, if precautions or not taken.
At RS Microwave, available computer simulation techniques are sufficiently sophisticated that accurate prediction of performance and dimensions minimizes the time required to develop and deliver such complex assemblies. Interconnection of sub-components or sub-modules within multiplexers is sometimes difficult, with parasitic lengths causing degradation of performance. Although the computer can predict these problems, sometimes the parasitics reach levels for which compensation cannot be effected. At RS Microwave, proprietary blind-mate interconnection of submodules is used to minimize both parasitic interconnections and spurious crosstalk. Thus, the physical structure, including all interactions, can be predicted accurately and the unacceptable interactions and crosstalk eliminated using the mechanical elegance and electrical isolation of blind-mate internal connections.
Multiplexer development is impacted heavily by network synthesis and computer simulation techniques. As it becomes possible to synthesize combinations of lumped, distributed and evanescent elements as well as predict and compensate their interactions, multiplexers will shrink in size, increase in order (number of channels) and display improved performance in insertion loss, isolation and bandwidth.
As discussed in the multiplexer section, at frequencies common to two adjacent channels of a reactively coupled, contiguous (crossover) multiplexer, the minimum loss that can be experienced is 3 db. At the common frequency, half the power will go into one channel and half into the other. If this loss is not acceptable, a price must be paid. This price is a loss of the ability to simultaneously view the outputs of both channels. In short, one must switch between the two channels. In most cases the data output from a wide band multiplexer channel must be processed through some digital data device. The time to perform such processing is usually much greater than the time required to switch between two channels. Thus, although simultaneity is lost, effectively it is maintained in the switched filter because one channel is being processed while the switching is occurring.
Another interesting feature of the switched filter is the relative ease with which higher order (more than three channels) structures may be constructed as compared to a multiplexer. The combination of blind-mate or drop-in filters, channelized microstrip interconnections, miniature switch modules and integrated drivers can form compact switched filter assemblies that will compete effectively with multiplexer techniques in wide band receiver applications. For higher power situations, the switches represent a power/switching speed limitation.
Switch-filter assemblies have characteristic implementation difficulties. These difficulties center around the problem of eliminating crosstalk between channels. Crosstalk occurs outside of the switch or filter, simply due to excess radiative or conductive coupling between the lines leading up or away from the switch or filter. The problem is usually made even more serious because conventional switched-filters employ drop-in switches and drop-in filters. Thus, the external line required to interconnect the switches and filters is at least 0.2” long. With filter and switch wall thickness of 0.100”, a total interconnection length of 0.400” results (Fig. SW-1). Perhaps equally as severe is the fact that the external interconnection line is not well-defined in terms of impedance or electrical length, and thus creates uncertainty in the simulation of the composite switch and filter response. At very low frequencies, such a short line will not radiate enough to cause problems. However, at frequencies as low as 100 MHz, the short line, an inefficient radiating element, radiates enough energy to adjacent regions within the assembly that circuit isolation is reduced to less than the intrinsic switch isolation. Thus, no matter what isolation properties are displayed by the switch, the crosstalk problem reduces channel-to-channel isolation, frequently below specification.
At RS Microwave, we have virtually eliminated the aforementioned problem, using our proprietary blind-mate interconnection technique. The switches are provided with blind mate shrouds, filters with proprietary but GPO-compatible blind-mate fingers. Thus, the filters “plug-in” to the switches. The total interconnection length, including filter and switch walls, does not exceed 0.260” (Fig. SW-2). No interconnecting line is required. Isolation is intrinsically at least 85 dB to 18 GHz, and thus the switch isolation is not compromised by interconnection problems. The interconnection is well-defined (i.e. 0.260” of 50 ohm line) and thus can be easily incorporated into simulation data. The reduced length of the interconnection (0.260” versus 0.400”) also helps to eliminate parasitic responses.
Switches are sometimes more expensive than passive filters, sometime display more insertion loss and certainly consume DC power. If filter networks can be substituted for switches, some reduction in loss, size and cost can ensue. In the cases where filter passbands are separated by some guardband, filters can be multiplexed at a common end and switched at the other end (Fig. SW-3). This application is useful in generation of a series of local oscillator tones (“comb-generator”), because the filter passbands are narrow and well-spaced. If the input is multiplexed and the output switched, the RF power at the switch has already been reduced by the filter insertion loss, and the harmonics and spurious products generated by the switch are consequently reduced.
The switched filter technology is becoming increasingly more affordable, as MMIC based FET switches increase in frequency range. For applications below 6 GHz, this FET based technology is the low power approach of choice (70841A-3, 61461A-3). It is likely that the FET circuitry will be practical at frequencies up to 18 GHz or so in the next few years. Presently, PIN diodes remain the main alternative to the FET approach for fast switching, and the only solid-state technology available for high power switched filters.
RS Microwave has developed many such high power assemblies (50703-4). Such assemblies frequently contain bandpass filtering or remove harmonics or to protect against lightning by providing a DC short circuit at the input to the assembly. PIN diode switches require more driving current than FET based switches, but can provide low loss and high isolation when properly combined with RS Microwave’s blind-mate approach to integration. For slower switching, it is still possible to select filters using mechanical relay type switches. Such relays switch in milliseconds, not microseconds, but the switches only contribute 0.1 to 0.3 dB insertion loss, rather than losses of 1 to 2 dB as found in most solid state switches. It is possible to combine filter and switch assemblies with gain blocks, however, to reduce or eliminate insertion loss, if noise figure and compression do not become major concerns. As technology improves, RS Microwave will continue to integrate the best switches and gain blocks with the best filters.
Cavities are resonant structures in which the operating mode can be coaxial or waveguide. Cavities serve the same purpose in microwave filters as do "tank" circuits in low frequency networks. Coaxial cavities employ a center conductor while waveguide cavities do not. Typically, above 2 or 3 GHz or inapplications where Q factors must be great (i.e. low loss or high power), waveguide cavities must be used. When the cavities are coupled to oneanother we form the equivalent of coupled tank circuit filters. At RS Microwave, we normally employ evanescent mode filters in place of coaxial cavities except in the case of tunable coaxial filters. The simplest form of waveguide coupled cavity filter consists of a length of waveguide divided by partitions into half-wavelength long rectangular cavities. Using single or multiple openings (called "irises") in these partitions, energy is "leaked" from one cavity to the next. The desired amount of leakage is determined from bandwidth, frequency or other network considerations on a theoretical and/or experimental basis. Such rectangular cavity filters can achieve passband widths of up to 10% or so with stopbands extending down to DC and up to 1.4 times center frequency.
It is also possible to form cavities with circular cross-sections and to iris couple them either "end-to-end" or through a common side wall. The latter case is useful for building mechanically tunable filters. Such circular cavities can operate in a variety of modes, with higher Q (less loss) but lower bandwidth (a few percent is maximum) than the rectangular types. As well, stopband widths are lower, extending only from 8% to 18% above the center frequency of the filter. Thus, such filters are utilized in low loss preselection applications and usually in conjunction with some additional filtering to enhance the stopband width. It is also possible to build coupled cavity filters in which each cavity is simultaneously resonant in two modes (dual mode filters) or in which each cavity is resonant in a different mode (multimode filters). The latter case is interesting because the modes can be selected to improve the overall loss characteristic while also increasing the stopband width. The stopband improvement occurs because the spurious resonant frequencies of the different possible operating modes do not coincide. Filters of the dual and multimode types are usually difficult to design because complex, non-cookbook iris structures are required. Our computer aided designs enable selection of optimum mode and iris combinations, with new experimental and theoretical information being constantly "fed-back" into the programs. As computer models are improved, the time to turn out complex filters decreases and the performance improves.