System-level analog design is a process largely dominated by heuristics. Given a set of specifications/requirements that describe the system to be realized, the selection of an optimal implementation architecture comes mainly out of experience. Usually, what is achieved is just a feasible point at the system level, while optimality is sought locally at the circuit level. The reason for this is the difficulty in the analog world of knowing if something is realizable without actually attempting to design the circuit. The number of effects to consider and their complex interrelations make this problem approachable only through the experience coming from past designs. One of the main problems is that optimization at the system level is really hard because of the difficulties in abstracting single block behaviors at the system level and in estimating their performances (and feasibility) without further knowledge of the implementation architecture and topology.
The basic idea of the proposed methodology is to extend the concept of platforms developed in the digital world to analog contexts. With an analog platform (AP)  we indicate an architectural resource to map analog functionalities during early system level exploration phases and constrain exploration to the feasible space of the considered implementation (or set of implementations). Therefore, an AP consists of a set of behavioral models and of matching performance models. Behavioral models are abstracted parameterized models of the analog functionality which introduce at the functional level a number of non-idealities attributable to the implementation and not intrinsic in the functionality, such as distortion, bandwidth limitations, noise, etc.
We propose applying the AP paradigm to design cases from the BWRC and extending the methodology to capture more accurate design constraints. The activity should start by encapsulating the block architectures required for the candidate transceiver with the AP abstraction. For each block, a set of behavioral models has to be defined and implemented in some executable form, e.g., AMS languages or Simulink/Matlab. Then, a characterization policy and a sampling strategy have to be set up to generate performance models. Particular attention has to be focused on the interfaces of single blocks so that loading effects are properly considered and modeled. For example, the input stage of the mixer has to be considered when extracting the performance figures for the driving LNA, otherwise the loading effect present in the real implementation would not be estimated. For each block, different architectures should be modeled and merged in the platform stack. This allows having a level of abstraction in the platform stack where, for example, two or more LNA topologies are represented by the same functional model and by the union of the respective performance models. When operating at this level, a platform instance selection intrinsically performs architecture selection choosing a performance set achieved by architecture a rather than architecture b or c. At the system level, system optimization will be carried out using a non-linear optimization technique based on behavioral simulations. Possible candidates are simulated annealing and genetic algorithms, which are expected to perform well given the limited number of optimization parameters available at the system level. The challenging design of the reconfigurable radio-mobile terminal should emphasize the strengths and/or weaknesses of APs in evaluating tradeoffs over a wide range of opportunities.