Integrated Noise Power Research Articles

Evaluation of contrast and denoising effects related to imaging parameters of compressed sensitivity encoding in contrast-enhanced magnetic resonance imaging.

To acquire reference data for setting an appropriate compressed sensitivity encoding (CS) for brain lesion detectability, the effects of contrast and noise on contrast-enhanced magnetic resonance imaging (MRI) were evaluated. Gadobutrol at various concentrations and manganese chloride tetrahydrate were used as a phantom. Various CS factors (0-10) and denoising levels (weak, medium, and strong) were assessed. The contrast amount decreased from CS7 in non-denoised images for 0.5-2mmol/L solutions but slightly decreased from CS7 with denoising. The noise amount significantly increased with an increasing CS factor. Generally, there was a significant difference in the denoising level and rate across all CS factors in the case of the 2 and 0mmol/L solutions. When the CS factor was increased without denoising, the integrated noise power spectrum (NPS) increased and decreased in the high-frequency and low-frequency areas, respectively. These data can be used to establish settings based on the degree of denoising.

Validation of a Digital Noise Power Integration Technique for Radiometric Clear Sky Attenuation Estimation at Q-Band

This article presents the validation of a digital signal processing technique that can be used to estimate radiometric sky noise, and hence atmospheric absorption, within existing digital receivers at little/no additional cost. To demonstrate this, a receiver was constructed that simultaneously records the beacon signal power from the ALPHASAT Aldo Paraboni technology demonstration payload, as well as the integrated noise power in the adjacent band. Calibration from the digital radiometer is performed using tip-curve calibration procedures. Atmospheric fading is then obtained by observing the beacon as well as the radiometric signals. This enables the comparison of fading obtained by the two techniques and provides a means to calibrate the received beacon power level to obtain total atmospheric attenuation. It is shown that for low levels of fading, up to a few dB, the two techniques provide good agreement. This approach can, therefore, provide a low-cost option for geostationary mm-wave satellite channel measurements in the low fading regime, which can be useful in the design and operation of the feeder links in emerging satcom systems.

Spin Torque Effect in Magnetic Low-Frequency Noise Due to Asymmetric Oscillation

In this letter, the influence of spin-transfer torque (STT) on magnetic read sensor noise in the test band of 1–200 MHz has been investigated experimentally using the macrospin model with STT term. The results reveal a dependence of the integrated noise power on the direction of bias current when an external field is applied. In a positive field direction, positive bias current produces stronger noise than negative current. In the presence of negative field direction, the relationship is reversed. The hypothesis of asymmetric oscillation noise modification by STT is confirmed by corresponding changes of a correlation between the signal in the test band, and amplitude fluctuations of thermally driven ferromagnetic resonance in the sensor free layer, both derived from high-bandwidth waveforms. The effect is a consequence of the spin torque impact on the effective damping constant, which adjusts the shape of the asymmetric oscillation spectrum in a test band.

Exploiting Inherent Error Resiliency of Deep Neural Networks to Achieve Extreme Energy Efficiency Through Mixed-Signal Neurons

Neuromorphic computing, inspired by the brain, promises extreme efficiency for certain classes of learning tasks, such as classification and pattern recognition. The performance and power consumption of neuromorphic computing depend heavily on the choice of the neuron architecture. Digital neurons (Dig-N) are conventionally known to be accurate and efficient at high speed while suffering from high leakage currents from a large number of transistors in a large design. On the other hand, analog/mixed-signal neurons (MS-Ns) are prone to noise, variability, and mismatch but can lead to extremely low-power designs. In this paper, we will analyze, compare, and contrast existing neuron architectures with a proposed MS-N in terms of performance, power, and noise, thereby demonstrating the applicability of the proposed MS-N for achieving extreme energy efficiency (femtojoule/multiply and accumulate or less). The proposed MS-N is implemented in 65-nm CMOS technology and exhibits $> 100\times $ better energy efficiency across all frequencies over two traditional Dig-Ns synthesized in the same technology node. We also demonstrate that the inherent error resiliency of a fully connected or even convolutional neural network can handle the noise as well as the manufacturing nonidealities of the MS-N up to certain degrees. Notably, a system-level implementation on CIFAR-10 data set exhibits a worst case increase in classification error by 2.1% when the integrated noise power in the bandwidth is $\sim 0.1 ~\mu $ V2, along with $\pm 3\sigma $ amount of variation and mismatch introduced in the transistor parameters for the proposed neuron with 8-bit precision.

Random telegraph signals with time dependent capture and emission probabilities: Analytical and numerical results

In this work, we study the noise spectra of random telegraph signals (RTS) with time dependent capture and emission probabilities. The capture and emission probabilities are made time dependent by cyclo-stationary excitation. Rigorous analytical calculations are performed to derive the model equations, and Monte Carlo simulations are performed to explore the noise behavior and validate the analytical model. Two study cases are explored: input signals modeled by periodic square wave and sinusoidal waves. We present clear situations where the integrated noise power is smaller than stationary case for a suitable choice of parameters in both cyclo-stationary inputs studied.

Low-frequency noise statistics for the breakdown characterization of ultrathin gate oxides

We have investigated the statistics of low-frequency noise in the tunneling current of ultrathin oxides (2.5–4nm) in metal-oxide-semiconductor capacitors as a function of the applied voltage stress. The statistical analysis includes (i) non-Gaussianity (nG), which is a measure of the degree of temporal correlation in the noise, and (ii) ratio of integrated noise power to the dc leakage current (R). The occurrence of high peaks in nG indicates the appearance of new percolation paths, and the subsequent conduction through these paths is indicated by R. Our results show that the nG and R characteristics are generic for the oxides of different thickness and growth quality and have the potential, in conjunction with leakage itself, of being used as a prognosticator of oxide reliability.

Novel analytical and numerical approach to modeling low-frequency noise in semiconductor devices

In this work, based on robust experimental results, we derive a novel theoretical probabilistic approach to model the low-frequency noise power S ( f ) in semiconductor devices. Using the proposed approach we obtained, analytically, the average of the integrated noise power W p = ∫ f L f H S ( f ) d f as a function of the frequency bandwidth [ f L , f H ] , showing that 〈 W p 〉 ∝ ln 2 ( f H / f L ) . The second moment and the relative error in W p are also calculated. A numerical fit for the relative error was performed, showing that lower and upper bound estimates to noise power depend only on the ratio f H / f L .

Percolation Studies in Synthetic Ferrimagnetic Recording Media

We have studied the percolation limit of synthetic ferrimagnetic media (SFM) by the measurement of noise power and third harmonic ratio and by imaging with a high-resolution magnetic force microscope (MFM). The noise power spectrum, which helps to find the onset of percolation, is significantly different in SFM structures compared to the conventional recording media. Properties of the stabilization layers, such as switching field distribution (SFD), are highly influential in deciding the integrated noise power spectrum. With suitable selection of high moment stabilization layers, the noise at higher densities, within the percolation density limit, can be substantially reduced. Third harmonic ratio studies indicate that the partial erasure occurs above /spl sim/900 kfci. Using a high-resolution high-Q MFM, it is possible to obtain stable and nonpercolating high density patterns up to 1000 kfci on media with tBr=0.22 memu/cm/sup 2/ and Hc/spl sim/4500 Oe. This results in practical designs of media for recording densities >150 Gbits/in/sup 2/.

Quantifying advanced tape medium noise

An investigation of the tape medium noise mechanism analyzed the magnetization fluctuation produced by an incremental dc erase process in a variety of modern metal-particle tape media. It examined the wavelength characteristics of the noise and compared the integrated noise during the reverse dc erase process with in situ remanent hysteresis curves. The integrated noise power followed the square of the first field derivative of the remanent hysteresis curve, (dM/dH)/sup 2/, showing that spatial fluctuation of the recording field inside the medium is a major source of the tape medium noise. It excluded spatial fluctuation of the medium coercivity as a possible noise source by comparing results from recording heads with two very different gap lengths. A quantitative analysis of the medium noise spectra at different erase currents indicated that interfacial roughness at the magnetic coating backside in some dual-layer media is a major source of medium noise. The conclusion: The main mechanism of the medium noise in advanced tape media is the head medium spacing fluctuation due to surface roughness.

Bulk noise processes and their correlation to structural imperfections in magnesium-doped p-type GaN grown on sapphire

We investigate the noise properties of p-type Mg-doped GaN using low frequency noise spectroscopy. The epitaxial GaN:Mg films were grown on a sapphire substrate by metalorganic chemical vapor deposition in different laboratories. Generation–recombination (g–r) noise and one-over-f (1/f) noise are observed for temperatures above 250 K. The magnitude of the 1/f noise exceeds the g-r noise magnitude for frequencies less than 30 Hz, and the 1/f noise level is characterized by high values of the Hooge parameter, α≈1–150, indicating a high level of structural imperfection. In addition, the integrated noise power spectral density divided by the voltage squared in the frequency range of 1 to 30 Hz, correlates strongly with the structural imperfection of the sample as measured from the asymmetric rocking curve (ω scan) full width at half maximum. The generation-recombination noise is related to a high concentration trap level with an activation energy of 120±25 meV and a repulsive barrier that is possibly associated with the Mg dopant.

Effect of magnetic anisotropy distribution in longitudinal thin film media

The effects of anisotropy axes distribution on the properties of longitudinal recording media are simulated using the micromagnetic modeling. The angle between the anisotropy axis and the preferred direction is assumed to follow the Gaussian distribution. Hysteresis loops are simulated with the field applied in the preferred direction and perpendicular to it. Orientation ratios (OR) are calculated for different anisotropy distribution. OR (coercivity) increased from 1.4 to 79.2 when standard deviation of anisotropy distribution σ decreased from 0.7 to 0.05. The effect of the anisotropy distribution on the recording performance is evaluated, too. The results show that narrowing the anisotropy distribution will increase the OR and decrease the transition noise power and transition length. It was found that a critical point occurred at σ=0.15, below which the OR can be maintained at a sufficiently high value and yet the integrated noise power can be kept low.

Computer simulations of low noise states in a high-power crossed-field amplifier

A large body of experimental data has been accumulated over the past 15 years or so on the remarkable ability of both magnetrons and CFA's to operate under certain conditions at noise levels comparable to those achieved in linear beam tubes. SAIC's MASK code, a 2 1/2 -D particle-in-cell code, has been able to compute total, integrated noise power to an accuracy of /spl plusmn/ a few dB in a high-power CFA, operating with a typical intra-pulse spectral noise density of /spl sim/47-50 dB/MHz. Under conditions that produced low noise (/spl sim/60-100 dB/MHz) in laboratory experiments, the MASK code has been, until now, unable to reproduce similar results. The present paper reports the first successful production of a very low noise state in a CFA simulation using the MASK code. The onset of this low noise state is quite sudden, appearing abruptly as the current is raised to a point near which the cathode operates as nearly emission limited. This behavior is similar to that seen in an experimentally observed transition between low noise and high noise operation in the SFD-266, a Varian [CPI] low noise CFA. Some comments are made concerning the nature of the noise as observed in the simulation and in the laboratory.

Transition position and amplitude fluctuation noise model for longitudinal thin film media

Integrated noise power in longitudinal thin film media increases with transition density. This phenomenon has been attributed to the medium noise being concentrated in the transition regions with the areas between transitions being relatively quiet. This paper presents a transition noise model for square wave recording on longitudinal thin film media considering two sources of noise: phase shift (or transition jitter) and amplitude. The transition jitter noise is due to random fluctuations in the transition positions, while the amplitude noise is due to random shape fluctuations in the transitions. It is assumed that random medium coercivity fluctuations cause transition position fluctuations. The random transition shape fluctuations are modeled by allowing the transition parameter a to be a random variable. These transition parameter variations are assumed to be caused by random medium magnetization fluctuations. Each written transition produces a demagnetizing field that affects the positions and shapes of subsequent transitions. The properties of both types of noise are discussed.

Transition noise model for longitudinal thin-film media

The integrated noise power in longitudinal thin-film media increases with transition density. This behavior has been attributed to transition noise predominating over the more conventional amplitude modulation noise. At low recording densities the total noise power increases linearly with the transition density. At higher recording densities the noise increases more rapidly. The departure from the linear noise power increase with transition density has been attributed to correlation between noise in adjacent transitions. A model is presented for the noise process in longitudinal thin-film media, assuming that each transition's position fluctuation is correlated to the previous transition's position fluctuation by the demagnetization field. The dominant effect of the demagnetization fields is to increase the inherent noise in each transition and broaden the transitions as the recording density increases. It is the inherent noise increase in each transition, rather than the contribution from the adjacent transition correlation noise power term, that explains the experimentally found nonlinear noise power increase at higher transition densities. The frequency dependence of the integrated noise power is shown to be a strong function of the transition shape.