6.5 Conclusions

We have examined some of the most widely used techniques for designing fully integrated continuous-time filters, and, of equal importance, methods by which these filters can be automatically tuned by on-chip circuitry. The frequency range over which these filters may be applied in wireless transceivers is extremely wide, ranging from audio frequencies to the RF signal frequency, which is normally in the UHF range, and with the general availability of deep-submicron IC processes is being extended to several gigahertz.

Integrated filter architectures can be divided into second-order cascade and multiple loop feedback types; a subset of the MLF filters includes a range of passive LC ladder simulation techniques. The second-order cascade offers simplicity of design with a modular circuit structure; however it possesses relatively high sensitivity compared to the MLF structures. Filters based on simulations of LC ladders are probably the most popular for integrated filter design. Their low sensitivity helps to offset the difficulties caused by loose integrated component tolerances. These may utilise component substitution and circuit transformation techniques to eliminate inductors from the prototype ladder design, or be based on ‘leap-frog’ structures which simulate ladder filter behaviour at a system level.

The OTA-C filter has important benefits for high-frequency operation. In this structure, all high-impedance nodes have grounded capacitors connected to them to define the filter response. This allows the inevitable stray capacitances to be absorbed into these tuning capacitors, without introducing additional unwanted parasitic poles into the response. The operational transconductance amplifiers operate in an open-loop mode; relatively low open-loop gain is required, which can be achieved by amplifier structures with wide bandwidth and low power consumption. The OTA is also well suited to electrical tuning of its transconductance, OTA-C filters have been successfully used for frequencies ranging from audio to several hundred megahertz. The main difficulties in designing OTA-C filters lie in the non-ideal behaviour of the OTAs used. The transistors making up the OTA are inherently non-linear, limiting the dynamic range of the filter. The finite output impedance of the transistors has the effect of reducing the maximum filter Q which can be achieved. Linearity and output impedance can be improved by resorting to more complex circuits; however these tend to have reduced bandwidths.

The chief advantage of the MOSFET-C technique is that good linearity can be readily achieved due to the high loop gain within the integrator circuits, and the distortion-cancelling properties of the balanced MOSFET-R structure. MOSFET-C filters have been very widely used as IF filters in receivers because of their low distortion capability. Typically, they are used from low frequency to tens of megahertz; the upper frequency limiting factor is the need for the amplifier gain–bandwidth product to be large in comparison with the operating frequency in order to achieve low distortion, and to suppress the effects of parasitic capacitance at the amplifier input and output. This usually means that, using a particular IC technology, higher operating frequency can be achieved using OTA-C techniques. However, with the very large bandwidths now achieved by submicron CMOS processes, this may not be a limiting factor, and successful MOSFET-C designs have been implemented for frequencies exceeding 100 MHz [21].

Active-LC filters are a recent development in integrated IC filter techniques. They promise some of the advantages of passive LC filters since the role of the active components in the filter is simply to replace energy dissipated in the losses of the passive components. Thus, potential advantages are low power consumption and good linearity at the highest frequencies. Several difficulties remain: the low Q and restricted range of values feasible for integrated spiral inductors in current IC processes, the difficulty of providing accurate compensation of losses over a wide frequency range and the difficulty of providing frequency and Q tuning. Nonetheless, active-LC filters with useful performance at frequencies of a few GHz have been described.

Tuning remains a critical issue for integrated filter designs. Virtually all filters employ some form of frequency tuning. With careful design, including the accurate modelling of parasitic effects, many designs operating at moderate frequencies (in the megahertz range), and with moderate Q (filter bandwidth of the same order as the centre frequency) have successfully achieved sufficient accuracy using relatively simple master–slave frequency tuning schemes without the need for Q tuning. For high-frequency and high-Q designs, Q tuning is required, due to the high sensitivity of Q to parasitic effects in the circuit. For such filters, the accuracy of matching between sections of the filter circuit may not be adequate to allow the use of master– slave techniques, so design of the tuning system may be much more difficult than the filter itself. Therefore, the filter design may not be feasible because provision of on-chip tuning is too difficult. One reason for the popularity of the low-IF superhet receiver architecture is the absence of requirements for narrowband filters. Improved tuning techniques for high-order filters operating at high frequency and Q filters remain an important research topic.

The field of integrated continuous-time filters is constantly evolving, being driven by progress in IC fabrication techniques. In the immediate future, the trend towards smaller geometry CMOS processes primarily designed for digital applications presents the filter designer with new problems. For example, reduced supply voltages limit the signal amplitudes that can be processed linearly within the filter circuit, without corresponding reduction in noise level. Thus, it is difficult to maintain dynamic range. MOSFETs with very short channel lengths exhibit low output resistance, making OTA design more difficult, especially since many circuit techniques for improving OTA performance are made impractical by the low supply voltage. There are also positive aspects; smaller geometries result in wider amplifier bandwidths, while copper interconnects and low K dielectrics should result in reduced parasitics and higher performance integrated inductors. Looking further ahead, new IC processes, using Si-Ge, or different semiconductor materials may become important, and require the development of new circuit techniques. Techniques such as MEMS and nanotechnology are already of commercial interest; if they become a feature of mainstream IC processes, a whole new range of components may become available to the integrated filter designer.