Hong-June Park
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Hong-June Park has a total of 88 co-authors. Here are the top co-authors by number of shared publications:
- Jae-Yoon Sim
Department of Electrical Engineering, Pohang University of Science and Technology, Pohang, South Korea
40 shared publications - Byungsub Kim
Department of Electrical Engineering, Pohang University of Science and Technology
26 shared publications - Hyunwoo Son
School of Electronic Engineering, Engineering Research Institute (ERI), Gyeongsang National University, Jinju-si, South Korea
3 shared publications - Kihwan Seong
K. Seong is with the Samsung Electronics, Hwaseong -si18448, South Korea
3 shared publications Min-Kyun Chae
Department of Electronic and Electrical Engineering, Pohang University of Science and Technology (POSTECH), Pohang, South Korea
3 shared publications- Minsoo Choi
Department of Electrical Engineering, Pohang University of Science and Technology, Pohang 790-784, South Korea
3 shared publications
Latest Publications
- Open Access
- 10 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 101698-101710
Area-Efficient Variable-Gain Delta-Sigma Modulator Achieving 11.5-dB Dynamic Range Improvement by Sharing Capacitors in the First-Stage Integrator
This paper presents an area-efficient variable-gain delta-sigma modulator (VGDSM) that shares the sampling, feedback, and integrating capacitors in the first-stage integrator while maintaining a constant sum of their capacitance values. By adjusting theVGDSMgain through sharing these capacitors, the proposed architecture increases the dynamic range (DR) compared to the fixed-gain delta-sigma modulator, while improving area efficiency compared to conventional VGDSM that use dummy sampling capacitors with fixed-valued feedback and integrating capacitors. For each VGDSM gain, the capacitances of the three capacitors were determined considering three design constraints related to area, noise, and linearity, respectively. Compared to the conventional VGDSM that includes dummy capacitors, this approach reduces the total capacitance of the first-stage integrator by 46.7%. Implemented in 180-nm CMOS technology, the modulator increased DR by 11.5 dB, to 100.4 dB. The active chip area is 0.301 mm², and with a 1.8-V supply, the power consumption is 732.8 μW, achieving a Schreier figure-of-merit of 175.7 dB.
- 10 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 101698-101710
Open AccessArea-Efficient Variable-Gain Delta-Sigma Modulator Achieving 11.5-dB Dynamic Range Improvement by Sharing Capacitors in the First-Stage Integrator
This paper presents an area-efficient variable-gain delta-sigma modulator (VGDSM) that shares the sampling, feedback, and integrating capacitors in the first-stage integrator while maintaining a constant sum of their capacitance values. By adjusting theVGDSMgain through sharing these capacitors, the proposed architecture increases the dynamic range (DR) compared to the fixed-gain delta-sigma modulator, while improving area efficiency compared to conventional VGDSM that use dummy sampling capacitors with fixed-valued feedback and integrating capacitors. For each VGDSM gain, the capacitances of the three capacitors were determined considering three design constraints related to area, noise, and linearity, respectively. Compared to the conventional VGDSM that includes dummy capacitors, this approach reduces the total capacitance of the first-stage integrator by 46.7%. Implemented in 180-nm CMOS technology, the modulator increased DR by 11.5 dB, to 100.4 dB. The active chip area is 0.301 mm², and with a 1.8-V supply, the power consumption is 732.8 μW, achieving a Schreier figure-of-merit of 175.7 dB.
- Open Access
- 17 June 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14  (12) , 5227
Speech Emotion Recognition under Noisy Environments with SNR Down to −6 dB Using Multi-Decoder Wave-U-Net
A speech emotion recognition (SER) model for noisy environments is proposed, by using four band-pass filtered speech waveforms as the model input instead of the simplified input features such as MFCC (Mel Frequency Cepstral Coefficients). The four waveforms retain the entire information of the original noisy speech while the simplified features keep only partial information of the noisy speech. The information reduction at the model input may cause the accuracy degradation under noisy environments. A normalized loss function is used for training to maintain the high-frequency details of the original noisy speech waveform. A multi-decoder Wave-U-Net model is used to perform the denoising operation and the Wave-U-Net output waveform is applied to an emotion classifier in this work. By this, the number of parameters is reduced to 2.8 M for inference from 4.2 M used for training. The Wave-U-Net model consists of an encoder, a 2-layer LSTM, six decoders, and skip-nets; out of the six decoders, four decoders are used for denoising four band-pass filtered waveforms, one decoder is used for denoising the pitch-related waveform, and one decoder is used to generate the emotion classifier input waveform. This work gives much less accuracy degradation than other SER works under noisy environments; compared to accuracy for the clean speech waveform, the accuracy degradation is 3.8% at 0 dB SNR in this work while it is larger than 15% in the other SER works. The accuracy degradation of this work at SNRs of 0 dB, −3 dB, and −6 dB is 3.8%, 5.2%, and 7.2%, respectively.
- 17 June 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14  (12) , 5227
Open AccessSpeech Emotion Recognition under Noisy Environments with SNR Down to −6 dB Using Multi-Decoder Wave-U-Net
A speech emotion recognition (SER) model for noisy environments is proposed, by using four band-pass filtered speech waveforms as the model input instead of the simplified input features such as MFCC (Mel Frequency Cepstral Coefficients). The four waveforms retain the entire information of the original noisy speech while the simplified features keep only partial information of the noisy speech. The information reduction at the model input may cause the accuracy degradation under noisy environments. A normalized loss function is used for training to maintain the high-frequency details of the original noisy speech waveform. A multi-decoder Wave-U-Net model is used to perform the denoising operation and the Wave-U-Net output waveform is applied to an emotion classifier in this work. By this, the number of parameters is reduced to 2.8 M for inference from 4.2 M used for training. The Wave-U-Net model consists of an encoder, a 2-layer LSTM, six decoders, and skip-nets; out of the six decoders, four decoders are used for denoising four band-pass filtered waveforms, one decoder is used for denoising the pitch-related waveform, and one decoder is used to generate the emotion classifier input waveform. This work gives much less accuracy degradation than other SER works under noisy environments; compared to accuracy for the clean speech waveform, the accuracy degradation is 3.8% at 0 dB SNR in this work while it is larger than 15% in the other SER works. The accuracy degradation of this work at SNRs of 0 dB, −3 dB, and −6 dB is 3.8%, 5.2%, and 7.2%, respectively.
- 21 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 71  (12) , 6010-6020
A PAM-4 Crosstalk Compensation Receiver Using an RC Circuit per Pin for 4×24 Gb/s/Pin Single-Ended DRAM Interface
A crosstalk compensation (XTC) receiver is proposed for a single-ended four-level pulse amplitude modulation (PAM-4) signaling DRAM interface. The receiver compensates for the crosstalk from two adjacent pin signals by using an RC circuit per pin. The transmitter of the test chip consists of a voltage-mode PAM-4 driver and a three-tap feed-forward equalizer. The XTC circuit at the test chip receiver consists of six blocks of a current-integrating differential amplifier per pin. The current integrating differential amplifier includes three differential pairs, one for its own pin data and the other two for adjacent pin data. The test chip, fabricated on a 28-nm CMOS low-power process, worked successfully at a data rate of 24 Gb/s per pin with four parallel 5-cm long microstrip lines on an FR-4 printed circuit board. The width was 14-mil and the center-to-center spacing was 35-mil. Compared to the previous crosstalk compensation scheme that uses three or more multi-pins, each with one RC circuit and two CR circuits per pin, the transmission coefficient at the receiver input pin was enhanced by 1.31 dB from –1.89 dB to –0.58 dB at the Nyquist frequency of 6 GHz. The XTC circuit opened the closed eyes of both the second and the third pin data among the four data pins of the test chip to 0.12 UI at 24 Gb/s per pin with PAM-4 signaling and bit error rate of 10 $^{-12}$ . The energy efficiency of the receiver test chip was 1.28 pJ/bit at the 1.1 V supply.
- 21 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 71  (12) , 6010-6020
A PAM-4 Crosstalk Compensation Receiver Using an RC Circuit per Pin for 4×24 Gb/s/Pin Single-Ended DRAM Interface
A crosstalk compensation (XTC) receiver is proposed for a single-ended four-level pulse amplitude modulation (PAM-4) signaling DRAM interface. The receiver compensates for the crosstalk from two adjacent pin signals by using an RC circuit per pin. The transmitter of the test chip consists of a voltage-mode PAM-4 driver and a three-tap feed-forward equalizer. The XTC circuit at the test chip receiver consists of six blocks of a current-integrating differential amplifier per pin. The current integrating differential amplifier includes three differential pairs, one for its own pin data and the other two for adjacent pin data. The test chip, fabricated on a 28-nm CMOS low-power process, worked successfully at a data rate of 24 Gb/s per pin with four parallel 5-cm long microstrip lines on an FR-4 printed circuit board. The width was 14-mil and the center-to-center spacing was 35-mil. Compared to the previous crosstalk compensation scheme that uses three or more multi-pins, each with one RC circuit and two CR circuits per pin, the transmission coefficient at the receiver input pin was enhanced by 1.31 dB from –1.89 dB to –0.58 dB at the Nyquist frequency of 6 GHz. The XTC circuit opened the closed eyes of both the second and the third pin data among the four data pins of the test chip to 0.12 UI at 24 Gb/s per pin with PAM-4 signaling and bit error rate of 10 $^{-12}$ . The energy efficiency of the receiver test chip was 1.28 pJ/bit at the 1.1 V supply.
- Open Access
- 4 February 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14  (3) , 1300
A Light-Weight Autoregressive CNN-Based Frame Level Transducer Decoder for End-to-End ASR
A convolutional neural network (CNN) transducer decoder was proposed to reduce the decoding time of an end-to-end automatic speech recognition (ASR) system while maintaining accuracy. The CNN of 177 k parameters and a kernel size of 6 generates the probabilities of the current token at the token level, at the token transition of the output token sequence. Two probabilities of the current token, one from the encoder and the other from the CNN are added to the frame level to reduce the decoding step to the number of input frames. An encoder composed of an 18-layer conformer was combined with the proposed decoder for training with the Librispeech data set. The forward-backward algorithm was used for training. The space and re-appearance tokens are added to the 300-word piece tokens to represent the token string. A space token appears at a frame between two words. A comparison with the autoregressive decoders such as transformer and RNN-T decoders demonstrates that this work provides comparable WERs with much less decoding time. A comparison with non-autoregressive decoders such as CTC indicates that this work enhanced WERs.
- 4 February 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14  (3) , 1300
Open AccessA Light-Weight Autoregressive CNN-Based Frame Level Transducer Decoder for End-to-End ASR
A convolutional neural network (CNN) transducer decoder was proposed to reduce the decoding time of an end-to-end automatic speech recognition (ASR) system while maintaining accuracy. The CNN of 177 k parameters and a kernel size of 6 generates the probabilities of the current token at the token level, at the token transition of the output token sequence. Two probabilities of the current token, one from the encoder and the other from the CNN are added to the frame level to reduce the decoding step to the number of input frames. An encoder composed of an 18-layer conformer was combined with the proposed decoder for training with the Librispeech data set. The forward-backward algorithm was used for training. The space and re-appearance tokens are added to the 300-word piece tokens to represent the token string. A space token appears at a frame between two words. A comparison with the autoregressive decoders such as transformer and RNN-T decoders demonstrates that this work provides comparable WERs with much less decoding time. A comparison with non-autoregressive decoders such as CTC indicates that this work enhanced WERs.
- Open Access
- 9 June 2022
- Published by MDPI AG in Electronics
- Vol. 11  (12) , 1831
A Low-Latency Streaming On-Device Automatic Speech Recognition System Using a CNN Acoustic Model on FPGA and a Language Model on Smartphone
This paper presents a low-latency streaming on-device automatic speech recognition system for inference. It consists of a hardware acoustic model implemented in a field-programmable gate array, coupled with a software language model running on a smartphone. The smartphone works as the master of the automatic speech recognition system and runs a three-gram language model on the acoustic model output to increase accuracy. The smartphone calculates and sends the Mel-spectrogram of an audio stream with 80 ms unit input from the built-in microphone of the smartphone to the field-programmable gate array every 80 ms. After ~35 ms, the field-programmable gate array sends the calculated word-piece probability to the smartphone, which runs the language model and generates the text output on the smartphone display. The worst-case latency from the audio-stream start time to the text output time was measured as 125.5 ms. The real-time factor is 0.57. The hardware acoustic model is derived from a time-depth-separable convolutional neural network model by reducing the number of weights from 115 M to 9.3 M to decrease the number of multiply-and-accumulate operations by two orders of magnitude. Additionally, the unit input length is reduced from 1000 ms to 80 ms, and to minimize the latency, no future data are used. The hardware acoustic model uses an instruction-based architecture that supports any sequence of convolutional neural network, residual network, layer normalization, and rectified linear unit operations. For the LibriSpeech test-clean dataset, the word error rate of the hardware acoustic model was 13.2% and for the language model, it was 9.1%. These numbers were degraded by 3.4% and 3.2% from the original convolutional neural network software model due to the reduced number of weights and the lowering of the floating-point precision from 32 to 16 bit. The automatic speech recognition system has been demonstrated successfully in real application scenarios.
- 9 June 2022
- Published by MDPI AG in Electronics
- Vol. 11  (12) , 1831
Open AccessA Low-Latency Streaming On-Device Automatic Speech Recognition System Using a CNN Acoustic Model on FPGA and a Language Model on Smartphone
This paper presents a low-latency streaming on-device automatic speech recognition system for inference. It consists of a hardware acoustic model implemented in a field-programmable gate array, coupled with a software language model running on a smartphone. The smartphone works as the master of the automatic speech recognition system and runs a three-gram language model on the acoustic model output to increase accuracy. The smartphone calculates and sends the Mel-spectrogram of an audio stream with 80 ms unit input from the built-in microphone of the smartphone to the field-programmable gate array every 80 ms. After ~35 ms, the field-programmable gate array sends the calculated word-piece probability to the smartphone, which runs the language model and generates the text output on the smartphone display. The worst-case latency from the audio-stream start time to the text output time was measured as 125.5 ms. The real-time factor is 0.57. The hardware acoustic model is derived from a time-depth-separable convolutional neural network model by reducing the number of weights from 115 M to 9.3 M to decrease the number of multiply-and-accumulate operations by two orders of magnitude. Additionally, the unit input length is reduced from 1000 ms to 80 ms, and to minimize the latency, no future data are used. The hardware acoustic model uses an instruction-based architecture that supports any sequence of convolutional neural network, residual network, layer normalization, and rectified linear unit operations. For the LibriSpeech test-clean dataset, the word error rate of the hardware acoustic model was 13.2% and for the language model, it was 9.1%. These numbers were degraded by 3.4% and 3.2% from the original convolutional neural network software model due to the reduced number of weights and the lowering of the floating-point precision from 32 to 16 bit. The automatic speech recognition system has been demonstrated successfully in real application scenarios.
- Open Access
- 24 May 2022
- Published by MDPI AG in Applied Sciences
- Vol. 12  (11) , 5300
A Convolution-Neural-Network Feedforward Active-Noise-Cancellation System on FPGA for In-Ear Headphone
A real-time streaming feedforward active-noise-cancellation (ANC) system for an in-ear headphone was demonstrated in a real application scenario, by implementing a 10-layer dilated convolutional-neural-network (CNN) on a field programmable gate array (FPGA). A 16 × 16 systolic array was used in the FPGA, to speed up the model computation. The system latency was 170.6 s, at the system clock frequency of 120 MHz. The CNN model used 3232 parameters. Due to the large input receptive field, of 327 ms, this work achieved total power reduction, of 14.8 dB and 14.3 dB at the noise incident direction of 0 and 90, respectively, and the noise attenuation bandwidth was 2000 Hz at both angles; all results were superior to those of the conventional FxLMS algorithm.- 24 May 2022
- Published by MDPI AG in Applied Sciences
- Vol. 12  (11) , 5300
Open AccessA Convolution-Neural-Network Feedforward Active-Noise-Cancellation System on FPGA for In-Ear Headphone
A real-time streaming feedforward active-noise-cancellation (ANC) system for an in-ear headphone was demonstrated in a real application scenario, by implementing a 10-layer dilated convolutional-neural-network (CNN) on a field programmable gate array (FPGA). A 16 × 16 systolic array was used in the FPGA, to speed up the model computation. The system latency was 170.6 s, at the system clock frequency of 120 MHz. The CNN model used 3232 parameters. Due to the large input receptive field, of 327 ms, this work achieved total power reduction, of 14.8 dB and 14.3 dB at the noise incident direction of 0 and 90, respectively, and the noise attenuation bandwidth was 2000 Hz at both angles; all results were superior to those of the conventional FxLMS algorithm. - 18 January 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 68  (7) , 2409-2413
A Duo-Binary Transceiver With Time-Based Receiver and Voltage-Mode Time-Interleaved Mixing Transmitter for DRAM Interface
A duo-binary signaling has been applied to both transmitter and receiver for high-speed low-power DRAM interface. The transmitter consists of a half-rate voltage-mode time-interleaved mixing duo-binary driver and a 2-tap feed-forward equalizer. The voltage-mode driver is used in this brief, because it generates more linear output voltage than the current-mode driver for the duo-binary signaling at low supply voltage. A time-based receiver is used for high-speed operation at low supply voltage. A 1-tap look-ahead decision-feedback equalizer scheme was applied to the time-based receiver for the duo-binary decoding operation. The test chip fabricated in a 28 nm low-power CMOS process gives the energy efficiency of 0.41 pJ/b at 12 Gb/s with a 25 cm-long FR-4 micro-strip line and the supply voltage of 0.75V.
- 18 January 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 68  (7) , 2409-2413
A Duo-Binary Transceiver With Time-Based Receiver and Voltage-Mode Time-Interleaved Mixing Transmitter for DRAM Interface
A duo-binary signaling has been applied to both transmitter and receiver for high-speed low-power DRAM interface. The transmitter consists of a half-rate voltage-mode time-interleaved mixing duo-binary driver and a 2-tap feed-forward equalizer. The voltage-mode driver is used in this brief, because it generates more linear output voltage than the current-mode driver for the duo-binary signaling at low supply voltage. A time-based receiver is used for high-speed operation at low supply voltage. A 1-tap look-ahead decision-feedback equalizer scheme was applied to the time-based receiver for the duo-binary decoding operation. The test chip fabricated in a 28 nm low-power CMOS process gives the energy efficiency of 0.41 pJ/b at 12 Gb/s with a 25 cm-long FR-4 micro-strip line and the supply voltage of 0.75V.
- 3 December 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 68  (6) , 1862-1866
A DFE-Enhanced Phase-Difference Modulation Signaling for Multi-Drop Memory Interfaces
This paper presents a phase-difference modulation signaling enhanced by decision feedback equalization for multi-drop memory interfaces. The phase-difference modulation signaling enables efficient data communication via a multi-drop channel, but only if the stub lengths are specially engineered. In this work, the phase-difference modulation signaling is combined with decision feedback equalization to extend the applicable channel range. A test chip with a phase-difference-modulation transmitter and a decision feedback equalization receiver was fabricated in 65-nm CMOS technology. With a 2-tap decision feedback equalization, the test chip achieved the data rate of 7.8 Gb/s/pin via the both multi-drop channels with a long 10-cm stub and a short 2-cm stub, whereas the previous work without decision feedback equalization does not work with the 2-cm stub channel at 7.8Gb/s/pin.
- 3 December 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 68  (6) , 1862-1866
A DFE-Enhanced Phase-Difference Modulation Signaling for Multi-Drop Memory Interfaces
This paper presents a phase-difference modulation signaling enhanced by decision feedback equalization for multi-drop memory interfaces. The phase-difference modulation signaling enables efficient data communication via a multi-drop channel, but only if the stub lengths are specially engineered. In this work, the phase-difference modulation signaling is combined with decision feedback equalization to extend the applicable channel range. A test chip with a phase-difference-modulation transmitter and a decision feedback equalization receiver was fabricated in 65-nm CMOS technology. With a 2-tap decision feedback equalization, the test chip achieved the data rate of 7.8 Gb/s/pin via the both multi-drop channels with a long 10-cm stub and a short 2-cm stub, whereas the previous work without decision feedback equalization does not work with the 2-cm stub channel at 7.8Gb/s/pin.
- Open Access
- 30 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 198468-198481
A Body Channel Communication Technique Utilizing Decision Feedback Equalization
This article presents the body channel communication technique that first adopts decision feedback equalisation. For the first time, we characterised post-cursor intersymbol interference of a single-bit pulse response of a human body channel for decision feedback equalisation achieving the first body channel communication faster than 100 Mb/s. In the proposed technique, an 8-tap decision feedback equaliser is utilised at the receiver to compensate for intersymbol interference of the body channel. The experimental results show that the proposed prototype transceiver achieved the highest data rates of 150 Mb/s and 10 Mb/s along 20-cm and about 100-cm body channels, respectively. Compared to the previous best result measured for the same body-channel distance of about 100 cm, the data rate is improved by ten times, showing that the proposed technique can significantly increase the speed of body channel communication.
- 30 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 198468-198481
Open AccessA Body Channel Communication Technique Utilizing Decision Feedback Equalization
This article presents the body channel communication technique that first adopts decision feedback equalisation. For the first time, we characterised post-cursor intersymbol interference of a single-bit pulse response of a human body channel for decision feedback equalisation achieving the first body channel communication faster than 100 Mb/s. In the proposed technique, an 8-tap decision feedback equaliser is utilised at the receiver to compensate for intersymbol interference of the body channel. The experimental results show that the proposed prototype transceiver achieved the highest data rates of 150 Mb/s and 10 Mb/s along 20-cm and about 100-cm body channels, respectively. Compared to the previous best result measured for the same body-channel distance of about 100 cm, the data rate is improved by ten times, showing that the proposed technique can significantly increase the speed of body channel communication.
- 3 March 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 67  (12) , 2863-2867
An 18-Gb/s NRZ Transceiver With a Channel-Included 2-UI Impulse-Response Filtering FFE and 1-Tap DFE Compensating up to 32-dB Loss
This paper presents a channel-included filtering feed-forward equalizer (FFE) scheme suitable for data transmission through a heavily-dispersive lossy channel. The proposed scheme adjusts the tap weights so that the impulse response seen at the receiver has only a 2-unit interval (UI)-long impulse response with other parts negligible. The confinement of each bit response to a 2-UI period effectively achieves a lossless duobinary-like eye opening and enables data recovery with only a 1-tap decision-feedback equalizer (DFE). The proposed transceiver, implemented in 28-nm LP CMOS technology, achieved data recovery of 18 Gb/s up to 32-dB loss with only 3-tap FFE at the transmitter and 1-tap DFE at the receiver.
- 3 March 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 67  (12) , 2863-2867
An 18-Gb/s NRZ Transceiver With a Channel-Included 2-UI Impulse-Response Filtering FFE and 1-Tap DFE Compensating up to 32-dB Loss
This paper presents a channel-included filtering feed-forward equalizer (FFE) scheme suitable for data transmission through a heavily-dispersive lossy channel. The proposed scheme adjusts the tap weights so that the impulse response seen at the receiver has only a 2-unit interval (UI)-long impulse response with other parts negligible. The confinement of each bit response to a 2-UI period effectively achieves a lossless duobinary-like eye opening and enables data recovery with only a 1-tap decision-feedback equalizer (DFE). The proposed transceiver, implemented in 28-nm LP CMOS technology, achieved data recovery of 18 Gb/s up to 32-dB loss with only 3-tap FFE at the transmitter and 1-tap DFE at the receiver.