Front End Design

Case Study

An ultrasound system needed to operate at several frequencies below 35MHz. Additional frequencies would be required at a later date, and upgrades were to be possible either at the factory or in the field.

The front-end to a sophisticated digital processing engine had to deliver data in the base-band domain. One of the advanced processing modes required the data to have very low inherrent jitter.

Basic Architecture

There are several design options available for such a system front end:

  • Digitization can be sufficiently over-sampled to satisfy Nyquist, with down-conversion and most filtering performed in the digital domain.
  • Direct down-conversion with sub-sampled digitization in quadrature; some analog and some digital filtering would be required.
  • Down-conversion using external mixers, with relatively low digitization rates of the quadrature signals; analog filters would be required both before (band-pass) and after (low-pass) mixing.

The various signal source spectra and dynamic ranges, and the nature of digital processing techniques all have a bearing on the optimum solution. Analysis of these determined that base-band digitisation was certainly an acceptable and probably the optimum technique for this particular system.


Filters will affect the signals being subjected to digital processing. The various standard filters (Bessel, elliptical etc) have differing effects. The combination of band-pass, mixing and low-pass in the context of particular signal sources and post-processing methods need analysis.

A particular generic characteristic was determined to be suitable for all filters, using scaling to suit the various frequencies involved. LC filters could theoretically meet all these requirements.

Could high-order LC filters be made economically with acceptable accuracy? Further analysis established how available components could be used, and yet still avoiding the need for adjust-on-test. The ability of the system to perform some moderate self-calibration was essential here.

To design a particular filter, attention to each component is required, particularly inductors. Simulation has to use relatively sophisticated models to ensure accuracy. This has a bearing on the bill of materials.


The main digital processing required that signals were devoid of certain features, such as dc and very-low frequency offsets. Analysis of these led to special algorithms that were implemented in a pre-processing FPGA.