Analog implementation of ANN with inherent quadratic nonlinearity of the synapses

In real-life applications of multilayer neural networks, the scale of integration, processing speed, and manufacturability are of key importance. A simple analog-signal synapse model is implemented on a standard 0.35 /spl mu/m CMOS process requiring no floating-gate capability. A neural-matrix of 2176 analog current-mode synapses arranged in eight layers of 16 neurons with 16 inputs each is constructed for the purpose of a fingerprint feature extraction application. Synapse weights are stored on the analog storage capacitors, and synapse nonlinearity with respect to weight is investigated. The capability of the synapse to operate in feedforward and learning modes is studied and demonstrated. The effect of the synapse's inherent quadratic nonlinearity on learning convergence and on the optimization of vector direction is analyzed. Transistor-level analog simulations verify the hardware circuit. System-level MatLab simulations verify the synapse mathematical model. The conclusion reached is that the proposed implementation is very suitable for large-scale artificial neural networks - especially if on-chip integration with other products on a standard CMOS process is required.     

Inverted-F antennas with wideband match performance

Novel inverted-F antennas, double-wire (DW-IFA) and double-strip (DS-IFA) were developed for f₀ = 2.45 GHz, and a 0.8λ ₀ in size square ground plate. The antennas are distinctive for their wideband match performance, 12% (numerical) for DW-IFA and 18% (numerical)/17% (experimental) for DS-IFA, and radiation characteristics suitable for the mobile environment scenario     

Statistical-uncertainty-based adaptive filtering of lidar signals

An adaptive filter signal processing technique is developed to overcome the problem of Raman lidar water-vapor mixing ratio (the ratio of the water-vapor density to the dry-air density) with a highly variable statistical uncertainty that increases with decreasing photomultiplier-tube signal strength and masks the true desired water-vapor structure. The technique, applied to horizontal scans, assumes only statistical horizontal homogeneity. The result is a variable spatial resolution water-vapor signal with a constant variance out to a range limit set by a specified signal-to-noise ratio. The technique was applied to Raman water-vapor lidar data obtained at a coastal pier site together with in situ instruments located 320 m from the lidar. The micrometeorological humidity data were used to calibrate the ratio of the lidar gains of the H(2)O and the N(2) photomultiplier tubes and set the water-vapor mixing ratio variance for the adaptive filter. For the coastal experiment the effective limit of the lidar range was found to be approximately 200 m for a maximum noise-to-signal variance ratio of 0.1 with the implemented data-reduction procedure. The technique can be adapted to off-horizontal scans with a small reduction in the constraints and is also applicable to other remote-sensing devices that exhibit the same inherent range-dependent signal-to-noise ratio problem.     

Rectangular backfire antenna with dielectric surface-wave structure

A rectangular backfire antenna with a dielectric surface-wave structure is studied experimentally and theoretically. It consists basically of two plane rectangular reflectors and a dielectric structure as a rectangular-waveguide prolongation. The antenna construction is compared with those of ordinary dielectric and horn antennas, which are more than twice the length of the backfire antenna. Simple expressions for the far-field radiation pattern are derived in terms of the near-field amplitude distribution.     

Silicon nitride Micromesh Bolometer Array for Submillimeter Astrophysics

We present the design and performance of a feedhorn-coupled bolometer array intended for a sensitive 350-mum photometer camera. Silicon nitride micromesh absorbers minimize the suspended mass and heat capacity of the bolometers. The temperature transducers, neutron-transmutation-doped Ge thermistors, are attached to the absorber with In bump bonds. Vapor-deposited electrical leads address the thermistors and determine the thermal conductance of the bolometers. The bolometer array demonstrates a dark noise-equivalent power of 2.9 x 10(-17) W/ radicalHz and a mean heat capacity of 1.3 pJ/K at 390 mK. We measure the optical efficiency of the bolometer and feedhorn to be 0.45-0.65 by comparing the response to blackbody calibration sources. The bolometer array demonstrates theoretical noise performance arising from the photon and the phonon and Johnson noise, with photon noise dominant under the design background conditions. We measure the ratio of total noise to photon noise to be 1.21 under an absorbed optical power of 2.4 pW. Excess noise is negligible for audio frequencies as low as 30 mHz. We summarize the trade-offs between bare and feedhorn-coupled detectors and discuss the estimated performance limits of micromesh bolometers. The bolometer array demonstrates the sensitivity required for photon noise-limited performance from a spaceborne, passively cooled telescope.     

Lactational performance,enteric gas emissions,and plasma amino acid profile of dairy cows fed diets with soybean or canola meals included on an equal protein basis

This study investigated the effects of feeding solvent-extracted canola meal (CM), extruded soybean meal (ESBM), or solvent-extracted soybean meal (SSBM) on an equivalent crude protein basis on performance, plasma AA profiles, enteric gas emissions, milk fatty acids, and nutrient digestibility in lactating dairy cows. Fifteen Holstein cows (95 ± 20 d in milk) were used in a replicated 3 × 3 Latin square design experiment with 3 periods of 28 d each. Treatments were 3 diets containing 17.1% CM, 14.2% ESBM, or 13.6% SSBM (dry matter basis). Vegetable oil was added (canola oil for CM or soybean oil for SSBM) to equalize the ether extract concentration of the diets. Rumen-protected Met was supplemented targeting digestible Met supply of 2.2% of metabolizable protein in all diets. Canola meal increased dry matter intake (DMI) by 5.9 and 8.9% in comparison with ESBM and SSBM, respectively. Milk urea nitrogen was lowest in CM, followed by SSBM, and was highest for ESBM. No differences were observed in feed efficiency, energy-corrected milk yield, and milk composition or component yields among treatments. Cows fed CM emitted less enteric CH₄ per kg of DMI compared with both ESBM and SSBM, but CH₄ emission intensity (CH₄ per kg of energy-corrected milk) was similar among treatments. In summary, replacement of ESBM or SSBM with CM, on an equal crude protein basis, in the diet of lactating dairy cows enhanced DMI, but yields of energy-corrected milk and milk components and feed efficiency were similar among treatments.     

Proteolytic activity in ruminal fluid from cattle fed two levels of barley grain: a comparison of three methods of determination

The effects of proportion of concentrate in the ruminant diet and the effects of freezing ruminal content prior to assay on proteolytic activity in ruminal inoculum were evaluated using three analytical techniques. A novel approach for determining proteolytic activity (PA) of ruminal fluid utilising ¹⁵N‐labelled casein was compared with two published procedures. In a crossover experiment, four heifers were fed two isonitrogenous diets containing (dry matter basis) 50% barley silage, 45% rolled barley grain and 4% soybean meal (medium‐grain diet, MG) or 8% barley silage, 89% rolled barley grain and 2% soybean meal (high‐grain diet, HG). Ruminal fluid was analysed either fresh or after having been frozen at ?40 °C for 45 days. Substrates utilised in measuring PA included ¹⁵N‐labelled casein (produced by infusing (¹⁵NH)₂SO₄ into the rumen of a lactating dairy cow), ¹⁴C‐labelled casein and azocasein. Incubations were conducted in 0.2 M phosphate buffer (pH 6.8) for 20 min at 39 °C. In the ¹⁵N‐casein incubations, PA was estimated as (i) N soluble in 5% trichloroacetic acid (TCASN), (ii) N soluble in 5% TCA corrected for microbial N uptake (TCAMICR) and (iii) N depleted from the soluble protein N pool (SPR). In the ¹⁴C‐casein incubations, PA was measured as TCA‐soluble radioactivity; in the azocasein method it was measured as dye released during incubation. Across treatments the highest (P < 0.001) proteolytic activity was measured by the SPR method, followed by TCAMICR, TCASN and ¹⁴C‐casein. The lowest activity was recorded using the azocasein method. Within the ¹⁵N‐ and ¹⁴C‐casein methods, PA in previously frozen ruminal fluid was higher (P < 0.05) with the HG diet than with the MG diet, and higher (P < 0.05) in previously frozen fluid than in inoculum processed fresh. This study demonstrates that increasing the proportion of grain in the diets of cattle and freezing the ruminal inoculum both increase proteolytic activity measured in ruminal fluid. The proposed ¹⁵N‐casein method yielded higher proteolytic activity values than the ¹⁴C‐casein method. Copyright © 2002 Society of Chemical Industry. Contributions of A N Hristov, T A McAllister and Z Xu. © Minister of Public Works and Government Services, Canada 2002.     

Effect of sodium laurate on ruminal fermentation and utilization of ruminal ammonia nitrogen for milk protein synthesis in dairy cows

A crossover design trial with 4 ruminally and duodenally cannulated lactating dairy cows was conducted to study the effect of sodium laurate on ruminal fermentation, nutrient digestibility, and milk yield and composition. The daily dose of sodium laurate (0, control or 240 g/cow, LA) was divided in 2 equal portions and introduced directly into the rumen through the cannula before feedings. Ruminal samples (29 in 114 h) were analyzed for fermentation variables and protozoal counts. Sodium laurate had no effect on ruminal pH and total and individual volatile fatty acids concentrations. Ruminal ammonia concentration, ammonia N pool size, and the irreversible loss of ammonia N were unaffected by treatment. Compared to control, protozoal counts were reduced by 91% by LA. Carboxymethylcellulase and xylanase activities of ruminal fluid were decreased (by 40 and 36%, respectively), and amylase activity was not affected by LA compared with control. Flow of microbial N to the duodenum was reduced by LA. Dry matter intake and apparent total tract digestibility of dry matter, organic matter, crude protein, neutral detergent fiber, and acid detergent fiber were not different between the 2 treatments. Milk yield, fat-corrected milk yield, milk fat and protein concentrations and yields, and milk urea N content were not affected by treatment. Sodium laurate did not affect transfer of ruminal ammonia-15N into bacterial or milk protein. In conclusion, LA at approximately 0.3% of the rumen weight reduced ruminal protozoal population and had a negative effect on fibrolytic activities of ruminal fluid and microbial protein flow to the intestine. Treatment had no other significant effects on ruminal fermentation, total tract digestibility, or transfer of ruminal ammonia-15N into milk protein.