On Comparison of Constrained and Unconstrained Evolutions in Analogue Electronics on the Example of "LC" Low-Pass Filters

Abstract

SUMMARYThe Evolutionary Electronics refers to the design method of electronic circuits with the help of Evolutionary Algorithms. Over the years huge experience has been accumulated and tremendous results have been achieved in this field. Two obvious tendencies are prevailing in the area over designers to improve the performance of Evolutionary Algorithms. First of all, as with any solution-search-algorithm, the designers try to reduce the potential solution space in order to reach the optimum solution faster, putting some constrains onto search algorithm as well as onto potential solutions. At the same time, the second tendency of unconstraining the Evolutionary Algorithms in its search gives unpredictable breakthroughs in results. Enabling the evolution to optimize with more experimental parameters devoted to drive the evolution and adjusted previously manually, is one of an example where such kind of unconstraining takes place. The evolution with the maximum freedom of search can be addressed as unconstrained evolution. The unconstrained evolution has already been applied in the past towards the design of digital circuits, and extraordinary results have been obtained, including generation of circuits with smaller number of electronic components. Recently, the similar method has been introduced by authors of this paper towards the design of analogue circuits. The new algorithm has produced promising results in terms of quality of the circuits evolved and evolutionary resources required. It differed from constrained method by its simplicity and represented one of the first attempts to apply Evolutionary Strategy towards the analogue circuit design. In this paper both conventional constrained evolution and newly developed unconstrained evolution in analogue domain are compared in detail on the example of "LC" low-pass filter design. The unconstrained evolution demonstrates the superior behaviour over the constrained one and exceeds by quality of results the best filter evolved previously by 240%. The experimental results are presented along with detailed analysis. Also, the obtained results are compared in details with low-pass filters designed previously. key words: evolutionary electronics, unconstrained evolution, analogue passive low-pass filter

Synthesis of Time-to-Amplitude Converter by Mean CoeVolution with Adaptive Parameters

“…In earlier works, we already had reported about some aspects of DM and adaptation [1,3]. In later works, we found that the system performs better at SRs 10 times lower than before (from 10% to 1%) [3].…”

“…In this work, we use relatively large populations (from 15,000 to 35,000 individuals evolving in parallel), each of which uses very low selection rates (SR) from 0.2% to 2% and aggregately the SR of the system reaches 0.048%. Furthermore, the proposed technique theoretically enables all the parallel evolutions "focus" on a single chromosome at one generation, which in our experimental case can bring up to 0.0006% SR 1 .…”

“…In this chapter, we present some of these failed attempts to enable a comparison. When applied toward less complicated circuits [1,3,7], our previous approach was able to solve successfully all the tasks set including low-pass filters, computational circuits, and 4-output Voltage distributor with superior functional and physical features. If earlier experiments [1] utilized 10%, the latter ones [3,7] applied 1% -10% SRs.…”

“…In recent years, evolutionary strategy (ES) has been found to be one of the most powerful evolutionary algorithms (EA) when applied toward the synthesis of electronic circuits [1][2][3][4]. Despite ES proving itself effective in solving the scalability problem and reaching great experimental results, the existing ES-based circuit synthesizers by their technique are far behind the state-of-theart EA theoretical developments.…”

“…To solve the problem, a recently developed ES-based system [1] is upgraded with a combination of the novel adaptive individual-level differentiated mutation (DM) and the winner-dominates-winner-cooperates (WDWC) coevolution strategy formed by means of parallel islandmodel evolution.…”