Pulse generator slashes the size and cost of 10 Gbit/s transmitters

High-capacity wireless systems are a great option for data transfer over distances of up to a few kilometers. If they can operate at 10 Gbit/s or more, they can help to address the rocketing demand for capacity in mobile phone and internet networks, which is partly fueled by the growth of uncompressed high-definition video. On top of this, wireless promises to be a popular alternative in locations where it is particularly challenging to lay fiber-optic cables for trunk-line data network systems. Mountainous regions and areas with rivers, roads or rail tracks all exacerbate the difficulties that are associated with build-out of buried fiber-optic networks.

The best frequencies for delivering data transmission at speeds in excess of 10 Gbit/s are found in the millimeter band. This region within the electromagnetic spectrum is rarely used for commercial applications, which makes it relatively easy to secure wide swaths of bandwidth. A subset of this band, the "radio window" of 70–100 GHz, is also subjected to a relatively low level of attenuation by atmospheric gases. This increases the data transmission range and improves the link s robustness.

W-band transmission
The short wavelengths that are associated with transmission in this radio window lead to equipment benefits. Millimeter-band radios can employ small hardware alongside high-gain antennas that produce a highly directional beam. This type of antenna reduces multi-path fading and interference with neighboring radios.

Despite these advantages, it is difficult to develop small, cost-effective equipment for transmitting in the 70–100 GHz range. This is because this type of unit currently requires the manufacture of several discrete single-purpose electronic components.

To address this weakness, our research team at Fujitsu and Fujitsu Laboratories, in Kanagawa, Japan, has embarked on the development of a radical form of impulse-radio technology. No oscillators or mixers are needed, and the transmitter can be built from just two active components: a pulse generator and an amplifier (figure 1). The pulse generator s emission corresponds to a data stream from a baseband block. RF signals, known as wavelets, are then created by passing this data stream through a band-pass filter. This filters a wide frequency spectrum of pulses to match the spectrum mask.

Our simple transmitter will only be a viable option for last-mile wireless if it can generate pulses with sufficient energy. Meeting this need requires very short pulses – the smaller the pulse s full-width at half-maximum (FWHM), the greater the power that is emitted at higher frequencies. Our application demands pulses with a FWHM of less than 10 ps, which we have realized with our InP HEMT technology. GaAs HEMT and sub-100 nm silicon MOS technologies were not employed because they have a weaker high-frequency performance and an inferior noise figure.

One distinguishing feature of our HEMT is its Y-shaped gate (figure 2). This architecture helps to prevent the top of the gate from peeling off and leads to a high gate-electrode fabrication yield. Conventional T-shaped gates are impaired by the low physical strength at the junction between a large top and a narrow stem. The Y-shaped gate is formed with a lift-off process that includes electron-beam lithography and the evaporation of a Ti/Pt/Au contact. The typical gate length is 0.13 µm.

Y-shaped gates have the additional advantage of being able to improve the uniformity and stability in an InP wafer. This enables more than 1000 HEMTs to be integrated in a single chip. Thinning of the Schottky barrier layer in each of these transistors enhances performance, including an increase in maximum transconductance and cut-off frequency to 1.5 S/mm and 220 GHz.

We used triple-layer gold-plated interconnections and benzocyclobutene dielectric films with a low dielectric constant to integrate the HEMTs into circuits. Metal-insulator-metal capacitors and NiCr thin-film resistors are also formed on the InP substrate.

Very simple digital-based pulse generators can be created with this technology. These consist of input and output buffers, delay control (DC) buffers and a pulse-generator core (figure 3a).

Short pulses are created in the core by adjusting the DC buffer delay times, which in turn tunes the overlapping time between input signals A and B (figure 3b). The pulse-generator core behaves like a logical NAND circuit and creates a pulse width that is nearly equal to the overlap.

Shorter pulses can be created by increasing the response speed of the pulse-generator core or the NAND circuit. This led us to develop a new, balanced NAND circuit (figure 3c) that can emit ultrashort pulses with a FWHM of just 7.6 ps (figure 4) and a peak-to-peak amplitude of 0.8 V.

We believe that these are the shortest pulses that have ever been generated from a semiconductor transistor. Spectral measurements reveal that the pulse generator can carry enough energy for frequencies above 100 GHz, which means that it s possible to build impulse-radio systems operating in the 70–100 GHz band.

Making a transmitter
We have followed this up by building a millimeter-band pulse transmitter (figure 5). This transmitter features a pulse generator, a 78–93 GHz band-pass filter and a W-band waveguide. Feeding 5 GHz clock signals into this module produced wavelets with a typical FWHM of 100 ps (figure 6). Spectral measurements confirmed that the wavelets had frequencies spanning the 78–93 GHz range, which almost corresponded to the transmission spectrum of the band-pass filter.

These results represent the world s first impulse radio transmitting at more than 10 Gbit/s in the millimeter band (in our case, 78–93 GHz). This new transmitter greatly simplifies the design of the millimeter-band emitters by eliminating the need for an oscillator, mixer and some other components. The upshot is a more compact package that is 70% smaller than its predecessors.

Our next aim will be to unite the transmitter with a receiver and conduct transmission testing with a fixed target. Field testing should follow.

Further reading
Y Kawano et al. 2006 IEEE Trans. Micro. Theory and Tech. 54 4489.
Y Nakasha et al. 2007 Ext. Abstr. of the 2007 SSDM 792.
Y Nakasha et al. 2008 International Microwave Symposium Digest 109.   

View pdf of article