III. Gm-C Filtering - Epublications - Université de Limoges

RF selectivity challenges
In this thesis, several topologies have been assessed. The purpose is to find which one
best suits the abacus created with the rejections specifications of the RF filter. It comes out
that the second order bandpass topology is the best. It actually needs few components which
make it an attractive structure for easy tunability. Moreover, it is also possible to tune the
quality factor Q of such a filter. Q describes the selectivity of the filter and is defined as the
ratio of the central frequency by the -3dB bandwidth. The higher the quality factor, the more
selective the filter.
Once the topology chosen, the literature has been assessed to find most appropriate
solutions. First, LC passive filters have first been studied as a comparison basis to active
solution studied further. A figure-of-merit has also been introduced to handle fair comparisons
with current state-of-the-art. As far as fully-active structures are concerned, the literature
mainly considers Gyrator-C and Gm-C filters. Gyrator-C filters consist in emulating an
inductive behaviour at RF frequencies by a gyrator, allowing the replacement of inductors by
active counterparts. Gm-C filters actually consist in synthesizing a transfer function by means
of integrators, and so of Gm-cells. For simple topologies such as the second order bandpass,
these two kinds of filter may lead to the same structure.
Gm-C filters
The studied Gm-C filter structure is based on an analogy with a parallel LC resonator
where the inductor is replaced by a gyrator. Tuning the gyrator value and tuning the
capacitance in parallel, a large tuning range is obtained compared to a passive LC filter. The
theoretical study of noise and linearity limitations leads to the choice of a moderate quality
factor since as soon as Q increases, RF performances are strongly degraded. This study also
demonstrates that high RF performances are required on all Gm-cells. It has been chosen to
use large transconductance values (2.5 and 10mS) in order to decrease the overall noise of the
filter. However, such transconductances have to be linearized to reach the linearity
specification set on the Gm-cell.
Several transconductor linearization techniques from the literature have been assessed.
Only two of them exhibit an interesting linearity versus noise trade-off: the Dynamic Source
Degeneration (DSD) which is also used in the literature for a TV reception RF filter [VI.1],
and Multiple Gated Transistors (MGTR) which show the most promising results at the Gmcell
level. Gm-C filters based on Gm-cells designed with these linearization techniques have
been simulated in 65nm CMOS technology. Table 33 summarizes the obtained performances
compared to literature results. The filter with MGTR-based Gm-cells presents the best
performances. For 6dB gain and a Q-factor close to 4, it exhibits an IIIP3 above 8dBm up to
450MHz, with a 16dB NF. However, this structure appears to be very sensitive to bias as well
as process and mismatch variations, where a robust solution is required for our applications.
That is why no test-chip has been realized based on this architecture.
- 173 -