This is the article that started it all, a proposal to transmit HD video as analogue over long distances of coaxial or twisted-pair cable. It was published in EDN magazine in December 2012. We originally called our proposal HD-CVI, but as this name was adopted by another company (and to a lower specification) we changed it to aCVi.
Transmitting HD video over RG-59 cable
High definition (HD) video cameras are becoming increasingly prevalent in security installations yet the existing cable infrastructure was designed to support only standard definition (SD) video. This article describes a method to transport HD video over >500m of low cost coaxial cable with little or no degradation.
There is an increasing need to be able to transmit high definition (HD) video over long cable runs, e.g. for security camera installations. In many cases the cable installation is pre-existing and uses low cost RG-59 cable but this cable is limited to low frequency use (<200MHz) for most applications; (at 1GHz its attenuation is 30dB/100m).
Existing methods to transmit HD video include separate analogue RGB/YPbPr, which requires three coaxial cables to transmit, or HD-SDI (and its derivative HD-CCTV), a serial digital transmission method which runs at a bit rate of 1.485MHz and can only achieve small distances with such cable.
Whilst analogue cable equalizers are available from some semiconductor manufacturers (e.g. Analog Devices and Intersil), they still require the transmission of 3 components (e.g. YPbPr) which involves the re-installation of additional cables which can add considerably to the total system cost of a security installation.
Other methods, such as compression of the video prior to transmission, (the IP camera), add considerable cost and power budget to each camera as well as limiting options in the post-processing of the image.
In this article we will discuss a method of transmitting HD video across this medium using a modified form of the well known NTSC analogue composite video standard to create the signal for transmission: we have christened the method HD-CVI, High Definition Composite Video Interface.
Distances of greater than 300m are achievable and in excess of 500m at 720p/60Hz with some small signal degradation. As with most analogue transmission methods, the signal degradation is ‘graceful’ with no sudden cut-off of the signal that is encountered with digital methods. HD-CVI is compatible with most HD video standards and also non-standard video, (e.g. from HD-CCD cameras).
HD-CVI is also backward compatible to standard definition NTSC/PAL transmissions whilst also offering a considerable transmission distance increase.
HD-CVI is also designed to support a back channel allowing control information to be embedded in the vertical blanking interval for information transfer between the receiver/transmitter; for example to control camera functions, and between the transmitter/receiver for embedding information about the transmitter signal source, standard or the camera environment. The back channel also permits automatic cable equalization.
RG-59 Cable Specification
Exact specifications vary across cable providers but an average of these will show an attenuation of approximately 2.4dB/100ft. at 50MHz and 3.5dB/100ft. at 100MHz, the frequency range of interest to us; (7.8db/100m and 11.5dB/100m respectively).
These figures do not take into account cable installation issues or other issues such as hum pickup. The HD-CVI interface has some degree of hum rejection through its pseudo-differential receiver input stage.
For the target distance specification of 300m and at the highest transmitted frequency (for 720p/60Hz = 73MHz) we are therefore required to accommodate approximately 3dB/100ft, (9.8dB/100m), or 29.5dB of loss.
The loss at 100kHz for a 300m cable run is 0.9dB.
A typical cable specification is shown in Figure 1.
Figure 1 Typical RG-59 Cable Specification from Hosiwell Technology Co. Ltd. At 1GHz the attenuation of 30dB/100m makes it totally unsuitable for transmitting HD-SDI at 1.485GHz.
HD-CVI Technical Overview
The following is a brief overview of the HD-CVI interface for the transmission of 720p/60Hz HD video. The parameters may be altered to adapt to other resolution requirements.
The basic concept of the HD-CVI interface is to build on the proven and reliable transport method of NTSC; (the advantages of PAL – v.v. multi-path reception – is not relevant to a cable system so the simpler NTSC standard is used as the model). NTSC transmissions are capable of being transmitted over more than 1km across RG-59 cable but the bandwidth is limited to 5MHz. NTSC also has chroma/luma crosstalk issues because of the interleaved nature of the signal that are difficult to resolve at the receiver end.
Because the cable system is a closed system it is only necessary for the transmitter and receiver to ‘understand’ each other and we can modify the basic NTSC method to suit HD transmissions.
The first thing to overcome is the bandwidth restrictions of the cable. HD 720p/60Hz transmission requires a luma bandwidth of 30MHz. Because we have only a single coaxial cable for the transport we have chosen to transmit luma and colour difference signals, (as opposed to component red, green blue), as the colour difference signals, because of the visual perception of the eye being less acute to colour, can be sent at half the bandwidth: i.e. 15MHz each. This is in accordance to the SMPTE-296M specification.
To further reduce the bandwidth of the transmission the colour difference signals are modulated onto a carrier in quadrature so they effectively use the same bandwidth. However, to avoid the signal recovery problems of NTSC, (and as we have no backward compatibility issues), the carrier is set above the luma bandwidth so there is no interference between the two signals. For 720p/60Hz transmission the carrier is at 58.559489MHz which ensures there is no beating with the horizontal or vertical scan frequencies; (the line frequency/carrier phase relationship of NTSC does not have to be adhered to because the chroma and luma are not interleaved). The carrier frequency is suitably far above the luma maximum frequency to allow inexpensive (in terms of silicon area) filters to be used in the transmitter and receiver.
The effective bandwidth of the complete signal is therefore approximately 15MHz + 58.6MHz or about 73MHz. This also sets the minimum sampling frequency at 2x 73MHz or 146MHz. For convenience we choose 148.5MHz as this is related to the 720p/60 SMPTE standard.
For 300m of RG-59 cable we can expect a 30dB loss at this frequency. However the synchronizing signals are at a much lower frequency where the loss is only about 1-2dB so reliable rastering of the received signal should always be assured.
To improve the signal to noise ratio (SNR) of the transmission pre-emphasis is used. The degree of pre-emphasis is programmable to allow for different cable lengths. The maximum pre-emphasis is set at 30dB for our evaluation and the frequency response is set to approximate the RG-59 cable loss.
A further improvement in the SNR is achieved through transmitting a peak to peak video level of 2V which maintains compatibility with any legacy SD equipment on the network and also allows common low-power 5V drivers to be used.
Beyond 300m the bandwidth will start to further fall off. The chroma signal will be the first affected by this being the highest frequency component. However automatic colour control in the receiver can maintain the colour saturation over a further 20dB signal attenuation, (to approximately 500m).
Fall off in the luma bandwidth can be compensated by a small boost in the receiver circuit. At extreme distances the signal will revert to monochrome as the chroma signal falls below the receiver’s range of compensation but synchronizing signals should be able to be received at greater than 1km distance. The luma bandwidth will be ‘gracefully’ reduced as the distance is increased.
Because of the similarity in the transmission method to NTSC both the transmitter and receiver can easily be made to accommodate conventional NTSC/PAL transmissions.
The HD-CVI Encoder
The following is a detailed description of the HD-CVI encoder (see Figure 2).
Figure 2 HD-CVI Transmitter Block Diagram . In this example the input is SMPTE-296M (720p/60Hz) component video at 74MHz. The output HD-CVI video at 148MHz can be transmitted across >300m of RG-59 cable with almost no degradation.
There are two clock inputs to the HD-CVI encoder, 74.25MHz (the SMPTE sampling frequency for the luma channel) and 148.5MHz. The 74MHz clock is used for the input data clocking and the 148MHz clock is used for the output clock timing. An additional enable is used for the Cb/Cr signals, (at 37.125MHz).
In addition two synchronizing inputs are required, HSync for the horizontal line input (45kHz) and VSync for the frame input (60Hz). From these inputs we create a composite sync pulse which complies with the SMPTE timing.
The HD-CVI sync waveform uses conventional syncs rather than tri-level syncs and with slower rise and fall times which helps to prevent ringing of the sync pulse edges with long cable runs and aid sync recovery. The rise and fall waveforms approximate a raised cosine shape over a period of 28 clocks. The shaped composite sync is then scaled and an offset added to avoid the negative clipping in the output amplifier.
The 10 bit luma data is scaled and added to the analogue composite sync output. The composite sync + luma signal is then interpolated from 74.25MHz to 148MHz in a 49 tap FIR filter. The filter has a pass-band of 30MHz and a stop band attenuation of > -50dB at 37MHz. The filter response is shown in Figure 3.
Figure 3 Luma Interpolator Filter response . The filter is flat to 30MHz allowing transmission of the full SMPTE specification luma bandwidth yet fully rejecting the chroma component ensuring no cross-colour effects.
The 10 bit B-Y and R-Y data is first scaled using the NTSC scaling parameters. The scaled U and V signals are then interpolated from 37.125MHz to 148MHz in a 63 tap FIR filter. The filter has a pass-band of 15MHz and a stop band attenuation of > -50dB at 20MHz. The filter response is shown in Figure 4.
Figure 4 Chroma Interpolator Filter Response . The 15MHz chroma bandwidth complies with the SMPTE-296M specification. By separating the luma and chroma channels we ensure there is no crosstalk and simplify the decoder design.
The subcarrier frequency used to modulate the chroma is generated using a 32 bit ratio counter clocked from the 148.5MHz clock.
The top 11 bits of this ratio counter (the phase word) are used by the demodulator to generate the sine and cosine waveforms.
The lower frequency of the luma filter cutoff is 37MHz and the cutoff frequency of the chroma filter is 20MHz. The minimum value for the subcarrier frequency to avoid crosstalk is therefore 57MHz. The subcarrier chosen for the 720p/60Hz standard is 58.559489MHz which ensures a small gap between chroma and luma and is also chosen to ensure there are no multiples of the horizontal or vertical scan frequencies which could cause beating effects.
The subcarrier phase word is used to address a ROM containing sine and cosine values. A sample of the sine waveform is added, after shaping, to the back porch of the video signal to synchronise the chroma demodulator of the receiver to provide an amplitude and phase reference to the receiver. This colour burst is blanked during the field pulse.
The sine and cosine values are multiplied, in turn, by the B-Y and R-Y. The resulting U.sin(2πF sc.t) and V.cos(2πF sc.t) data is added together to form the final chroma signal.
This chroma output is then passed through a notch filter centred on the subcarrier frequency. The notch ensures no harmonics generated by the modulator can interfere with the luma component. The response of the notch filter is shown in Figure 5.
Figure 5 The chroma modulator notch filter frequency response. This filter may prove unnecessary in the final evaluation but is designed to prevent any artifacts from the modulation process causing crosstalk in the luma channel.
The colour burst, chroma and luma with composite sync are added to create the complete HD-CVI output waveform. This waveform is then subjected to a variable degree of pre-emphasis with a maximum boost of +30dB at >50MHz. The pre-emphasis filter is a 5 tap FIR.
The degree of pre-emphasis is dependent on the cable length and is designed to approximately compensate for the loss of 300m of RG-59 cable. The response of the pre-emphasis filter is shown in Figure 6.
Figure 6 . The transmitter pre-emphasis filter frequency response. This filter approximates the high frequency attenuation of the coaxial cable and its effect is programmable depending on the length of the cable installation.
The 10-bit HD-CVI output from the pre-emphasis filter drives a high speed DAC, an Analog Devices (ADI) AD9705 for the evaluation platform.
The 1 volt nominal pk-pk output from the DAC differentially drives an ADI AD8051 high speed amplifier. This amplifier is set for a gain of x4 giving a peak to peak output of 4V which is then AC coupled into the cable via a 75Ω series termination. The AD8051 has a high current drive capability allowing it to fully swing the 4V into a 150Ω termination.
The output BNC is effectively floating although coupled to ground at high frequencies. This allows a pseudo differential mode for the received vertical interval data, (used, for example, for camera control).
At full pre-emphasis gain the sync outputs will be approximately 65mV pk-pk and about 1dB less than this at the receiving end after 300m of cable loss. Frequencies above 50MHz, subject to full pre-emphasis gain will be -30dB after 300m of cable or approximately 63mV pk-pk so for a typical installation our received frequency response will be essentially flat.
The HD-CVI input is AC coupled into a differential amplifier. The input is pseudo-differential at low frequencies which affords some hum rejection. The gain of this stage is -6dB; (ADI AD4830 amplifier – this amplifier also has input clamping to stop any electrical interference pickup destroying the front end).
The output from the input receiver is DC coupled into a voltage controlled amplifier, (ADI AD8337), which is able to apply up to +30dB of gain, sufficient for this evaluation. Another amplifier, or a cascaded design of this one, could be used for longer cable installations where more gain is required.
The output from this amplifier is then at a nominal amplitude for driving the ADC. This output is black level clamped using an analogue switch, the black level being clamped to 2.8V which is the operating point of the selected ADC analogue inputs.
The clamped signal is then buffered and provides the single ended drive into the ADC inputs. The ADC used for the evaluation is the ADI AD9430 (170MHz version).
The ADC is sampled at 148.5MHz. The output from the ADC is straight binary at 12 bits. Although 12 bits was used for the evaluation platform, a lower cost 10 bit ADC is expected to perform satisfactorily. Experiments are also being performed on dual ADCs, (sub 100MHz sample rates), operating them 180 degrees out of phase, as the dual ADCs are considerably lower cost than the single high speed ADC.
The following is a description of the HD-CVI decoder, (see Figure 7).
Figure 7 HD-CVI Decoder Block Diagram. Because the chroma is modulated onto a carrier above the luma signal there is no crosstalk and the decoder is considerably less complicated than a conventional NTSC video decoder with comb filter.
The HD-CVI input from the ADC is a straight binary, 12-bit input sampled at 148.5MHz. Analogue clamping prior to the ADC ensures the black level of the signal is at the midpoint of the ADC code, below that is sync, above that is video. There is no pedestal on the transmission.
The HD-CVI input is first low pass filtered to remove the chroma component. The filter is the same 49 tap FIR used in the transmitter, (see Figure 3). As the stop band of this filter is 37MHz we can decimate the output of the filter to 74.25MHz clock frequency.
A fixed offset is subtracted from the low pass filtered luma video such that the midpoint of the sync pulse is at value 0. Values 1-32 from the horizontal pixel counter address a look up table whose output coefficients form a FIR low pass filter to further reduce noise and subcarrier from the composite video. The coefficients are multiplied by the offset video and accumulated across the 32 samples, effectively being updated once per horizontal line. When the midpoint of the falling edge of the horizontal pulse is coincident with the midpoint of the FIR filter the accumulated result will be zero. When they are not coincident an error will be generated.
This error is filtered using a recursive filter (integrator) and proportional and integral terms are added to create an error word which is used to control a pulse width modulator output, which via a low pass filter produces an analogue control voltage to a voltage controlled oscillator (VCO). The resulting clock output from the VCO is the nominal 148.5MHz clock into the receiver module.
The horizontal pixel counter is used by the SPG, (sync pulse generator), to provide the horizontal timing pulses required by the decoder, including the black level clamp pulse to the analogue front end and the burst gate pulse for the demodulator.
The vertical field pulses are recovered by using a digital integrator on the sliced composite video.
A free-running subcarrier frequency is generated using the same method employed in the transmitter. The free-running frequency of the subcarrier is 58.559489MHz. The top 11 bits of this ratio counter (the phase word) are used by the demodulator to generate the sine and cosine waveforms.
For the demodulation to correctly operate the generated subcarrier must be frequency and phase locked to the HD-CVI video subcarrier which is done by measuring the amplitude of the demodulated and low pass filtered V output during the colour burst. If the frequency and phase of the free-running subcarrier and the colour burst are the same then this error will be zero. 32 samples of the V output waveform are taken during the burst pulse; the burst gate pulse from the SPG being used for this purpose. The seed word is then modified using the phase error signal until the input colour burst and the ratio counter are phase locked.
The HD-CVI chroma signal is originally generated as follows:
When the burst lock loop (BLO) is in lock, the frequency and phase will be the same as when the signal was being modulated. Thus, multiplying the HD-CVI composite video by the sine and cosine of the same frequency and phase gives the following:
and for the V component:
The reconstructed sine and cosine waveforms are multiplied by the input 148.5MHz free-running composite video. The output of the sine channel is the demodulated U signal and the cosine channel is the demodulated V output.
The output of the demodulator also comprises twice subcarrier frequencies. The output is therefore low pass filtered using a 63 tap filter, the response for which is shown in Figure 4. The output of the filter is the clean ‘simple’ demodulated U and V.
The low pass filtered luma is conditioned by the processing amplifier. First the black level offset is subtracted from the luma signal to set the black level at zero. The luma is then amplified to provide a 960 code (10 bit) output for a 100% white input. The luma output is sampled at 74.25MHz.
The low pass filtered chroma outputs are amplified separately to provide a nominal 700mV output for a 100% colour bar input. These outputs are valid on the rising edge of the 74MHz clock when the Cmux (37MHz) enable signal is high.
The SPG also provides Vout (vertical) and Hout (horizontal) synchronizing outputs which can be used for later processing blocks or for a digital to analogue converter.
HD-CVI Evaluation Platform
An evaluation platform for the HD-CVI interface has been developed to prove the concept. A small transmitter and receiver module have been developed and the encoder and decoder have been written in Verilog and compiled for an Altera Cyclone III FPGA (EP3C25).
The interface has proven reliable over 300m of RG-59 cable with a frequency response to 30MHz ± 0.2dB being achieved.
Figure 8 This is a screen capture of the output of the HD-CVI receiver. It shows a 30MHz sweep waveform after transmission across 300m of RG-59 cable.
Further work is proceeding on lowering the BOM cost, increasing the cable distance and transmitting 1080p/60Hz video.
I would be the first to complain against the proliferation of unnecessary standards or worse, the abuse of existing standards.
However, the increasing use of HD video cameras in the surveillance market has created a problem for the transmission of the same standards.
A large number of buildings have already been cabled for security, but using RG-59 cable with the expectation those cameras would be standard definition and existing methods of transmitting the video uncompressed are not compatible with these installations. Even for future installations either multiple cables or more expensive cable/repeaters will be required for the transmission of HD video, adding considerably to the cost of the installation.
The interface described above allows the transmission of HD video using RG-59 cable over distances of greater than 300m with very little degradation. The transmitter adds minimal cost overhead to the camera whilst the receiver up cost is expected to be comparable to adding HD-SDI. These costs could of course be mitigated by the development of an ASIC.