The high free-space loss and the limited performance of CMOS active and passive components at high frequencies impose the application of beamforming to enable wireless communication in the mm-wave band. At the receiver side, beamforming improves the SNR and increases immunity against interfering signals. However, broadband implementation of beamforming at mm-wave frequencies in CMOS requires extensive routing of high-frequency signals over the CMOS substrate, yielding loss and matching problems. In order to minimize these problems, in combination with low area and power consumption, phase shifting in the LO path is an appealing candidate [1]. In this paper, a CMOS implementation of this technique, based on a widely tunable QVCO is presented. Each path achieves 30dB of gain and a minimum NF of 7.1dB, yielding a system NF of 4.1dB. The overall current draw is 54mA from a 1.2V supply. Additionally, a calibration procedure to mitigate the analog impairments imposed by the proposed implementation is demonstrated.The architecture of the integrated 2-path receiver front-end section is depicted in Fig. 9.3.1. Each path consists of an LNA, a direct-downconversion mixer and a phase selector. The QVCO provides quadrature phases of the LO signal, routed to the phase selectors in both paths. Digital control of the phase selectors determines the phase of the LO signal at the input of each mixer.The single-stage common-source inductively degenerated cascode LNA ( Fig. 9.3.1) features frequency tuning, provided by a small digitally controlled load varactor C D . This feature allows in situ detection of the frequency where the LNA gain peaks, by performing an RF frequency sweep for both settings. The center frequency can be identified as the frequency where both receiver gain characteristics intersect. A 4-bit digital control on the size of the cascode transistor M 2 implements the gain selection in the LNA. A single-balanced Gilbert mixer with resistive load realizes the downconversion. It uses an inductor to tune out the capacitance at the drain of the RF input transistor [2]. The mixer is followed by a two-stage buffer, from which the last stage drives the 50Ω input impedance of the measurement equipment. The QVCO consists of two cross-coupled LC-VCOs, locked in quadrature by parallel coupling [3]. The 4 phases of the QVCO are applied to the 4 branches in the phase selector ( Fig. 9.3.1) that share a resonant load. The biasing of each branch is controlled separately, allowing phase manipulation of the signal that drives the LO input of the mixer. Our test-bench allows continuous beamsteering resolution, but in practice this resolution will be limited by the resolution of the integrated DAC.Implementation of broadband mm-wave multiple antenna systems in CMOS is complicated due to the high frequency signal loss and matching issues. Our layout approach to mitigate these problems is shown in Fig. 9.3.7. Most inductive components have been implemented as lumped components to optimize their area and boost their quality factor. Caref...

No abstract

In this paper we introduce the backward N-gram language model (LM) scores as a confidence measure in large vocabulary continuous speech recognition.Contrary to a forward N-gram LM, in which the probability of a word is dependent on the preceding words, a word in a backward N-gram LM is predicted based on the following words only. So the backward LM is a model for sentences read from the end to the beginning.We show on the benchmark 20k word Wall Street Journal recognition task that the backward LM scores contain information for the confidence measure that is complementary to the information in forward LM scores. The normalised cross entropy metric for confidence measures increases significantly from 18.5% to 23.3% when backward LM scores are added to a confidence measure which includes the forward LM scores.

Background Therapeutic drug monitoring (TDM) is a diagnostic tool in the monitoring of anti-TNF therapies. Yet, the benefit for TDM of new biologics such as ustekinumab (USK) is still controversial in real-world experiences. Methods This monocentric retrospective study aims to correlate USK trough levels (TLs) with clinical and endoscopic data. All patients have given written consent to the Biobank (B2011/005). Endoscopic disease was defined as quiescent in absence of endoscopic lesions, mild disease in presence of few superficial ulcerations, moderate in presence of several ulcers and severe in presence of numerous deep ulcers and/or inflammatory stenosis. 313 serum USK samples from 67 Crohn’s disease patients were used to measure USK TL (USK ELISA, apDia) while 88 samples (at week 16, and before and after optimisation) were used to measure anti-drug antibody (ADA), using a drug-tolerant affinity capture elution anti-ustekinumab assay Results Demographic and baseline data of our population are presented in Table 1. The median follow-up was 73 weeks (IQR 39–92). An optimisation due to loss of response was required in 44.8% of patients (n = 30) after a median time of 38 weeks (IQR 24–55). To evaluate the drug efficacy, an endoscopy was performed in 61% of cases at a median time of 35 weeks (IQR 27–47). TLs were 5.2 µg/ml (IQR 2.1–8.8), 1.7 µg/ml (IQR 0.3–4.3) and 2.6 µg/ml (IQR 0.6–4.1) at week 8, 16 and 24, respectively. TLs at week 8 were correlated to the induction IV dose administrated (r = 0.3, p = 0.03). At week 16, low TLs were associated with higher endoscopic activity in the follow-up (p = 0.02), although this was not the case at week 8 (p = 0.5) (Figure 1). Patients not requiring an optimisation had higher TLs in maintenance than patients requiring optimisation (2.45 µg/ml (IQR 1.3–4.4) vs. 1.15 µg/ml (IQR 0.1–2.24), p = 0.008). Obviously, optimisation significantly increased TLs (1.15 µg/ml (IQR 0.1–2.24) vs. 6.6 µg/ml (IQR 2.3–11.3), p < 0.001). ADA were undetectable in all the measured samples in maintenance. Conclusion This real-world experience confirms a drug exposure-endoscopic response relationship. Week 16 seems to be an appropriate time point to monitor drug exposure. Earlier USK TLs, at week 8, appear less valuable to be monitored due to the influence of initial IV dose. The absence of immunogenicity suggests that it is not a key driver in the loss of response.