-- -- Copyright 2021 Ettus Research, a National Instruments Brand -- -- SPDX-License-Identifier: LGPL-3.0-or-later -- -- Module: adc_gearbox_2x4 -- -- Description: -- -- Gearbox to expand the data width from 2 SPC to 4 SPC. -- library IEEE; use IEEE.std_logic_1164.all; use IEEE.numeric_std.all; entity adc_gearbox_2x4 is port( Clk1x : in std_logic; Clk3x : in std_logic; -- Resets with synchronous de-assertion. ac1Reset_n : in std_logic; ac3Reset_n : in std_logic; -- Data packing: [Q1,I1,Q0,I0] (I in LSBs). c3DataIn : in std_logic_vector(95 downto 0); c3DataValidIn : in std_logic; -- Data packing: [Q3,I3,Q2,I2,Q1,I1,Q0,I0] (I in LSBs). c1DataOut : out std_logic_vector(191 downto 0); c1DataValidOut : out std_logic ); end adc_gearbox_2x4; architecture RTL of adc_gearbox_2x4 is signal c1DataValidInDly, c3DataValidInDly : std_logic_vector(3 downto 0) := (others => '0'); subtype Word_t is std_logic_vector(95 downto 0); type Words_t is array(natural range<>) of Word_t; signal c3DataInDly, c1DataInDly : Words_t(3 downto 0); begin -- Pipeline input data. We will need four pipeline stages to account for the -- three possible Clk1x and Clk3x phases and the nature of data packing done -- in the DDC filter. The DDC asserts data valid for two clock cycles and -- de-asserted for one clock cycle. This requires us to have shift register -- that is 4 sample words (each sample word is 2 SPC) deep. InputValidPipeline: process(Clk3x, ac3Reset_n) begin if ac3Reset_n = '0' then c3DataValidInDly <= (others => '0'); -- These registers are on the falling edge to prevent a hold violation at -- the input to the following Clk1x FF (which may arrive late when more -- heavily loaded than Clk3x) elsif falling_edge(Clk3x) then c3DataValidInDly <= c3DataValidInDly(c3DataValidInDly'left-1 downto 0) & c3DataValidIn; end if; end process; InputDataPipeline: process(Clk3x) begin -- These registers are on the falling edge to prevent a hold violation at -- the input to the following Clk1x FF (which may arrive late when more -- heavily loaded than Clk3x). if falling_edge(Clk3x) then c3DataInDly <= c3DataInDly(c3DataInDly'high-1 downto 0) & c3DataIn; end if; end process InputDataPipeline; -- Data valid clock crossing from Clk3x to Clk1x Clk3xToClk1xValidCrossing: process(Clk1x, ac1Reset_n) begin if ac1Reset_n = '0' then c1DataValidInDly <= (others => '0'); elsif rising_edge(Clk1x) then c1DataValidInDly <= c3DataValidInDly; end if; end process; -- Data clock crossing from Clk3x to Clk1x Clk3xToClk1xDataCrossing: process(Clk1x) begin if rising_edge(Clk1x) then c1DataInDly <= c3DataInDly; end if; end process; ----------------------------------------------------------------------------- -- -- p0 p1 p2 p0 -- Clk3x _______/ŻŻŻŻŻŻŻ\_______/ŻŻŻŻŻŻŻ\_______/ŻŻŻŻŻŻŻ\_______/ŻŻŻ -- -- Clk1x _______/ŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻ\_______________________/ŻŻŻ -- -- c3DataValidIn _/ŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻŻ\_______________/ŻŻŻŻŻŻŻŻŻŻŻŻŻŻ -- -- This gearbox connect the DDC filter output to the remaining RX data path. -- For efficient use of DSP slices we run the DDC at 3x clock rate. Both -- Clk3x and Clk1x are sourced from the same PLL and is phase locked as shown -- in the above timing diagram. The output of DDC filter is asserted for two -- clock cycles and is de-asserted for one clock cycle. The remaining part of -- the design cannot run at 3x clock rate. So, we increase the number of -- samples per clock cycle and decrease the clock frequency to 1x. Depending -- upon the pipeline delay through the filter and RF section, the phase of -- data valid assertion could be on either p0, p1, or p2 edge. And depending -- upon the phase, data packing to Clk1x domain will vary. Since there are -- three possible phase, we will need three different data packing options. -- -- Data packing is done by looking for two consecutive ones in the data valid -- shift register (c1DataValidInDly).This pattern can be used only because of -- the way output data is packed in the filter. If we see two consecutive -- ones, then we know that we have enough data to be packed for the output of -- this gearbox. This is because, we need two Clk3x cycles of 2 SPC data to -- pack a 4 SPC data output on Clk1x. The location of two consecutive ones in -- the data valid shift register will provide the location of valid data in -- data shift register (c1DataInDly). DataPacker: process(Clk1x) begin if rising_edge(Clk1x) then -- Data valid is asserted when both Clk1x and Clk3x are phase aligned -- (p0). In this case, c1DataValidInDly will have consecutive ones in -- index 1 and 2. c1DataValidOut <= c1DataValidInDly(1) and c1DataValidInDly(2); c1DataOut <= c1DataInDly(1) & c1DataInDly(2); -- Data valid asserted on phase p1. if c1DataValidInDly(1 downto 0) = "11" then c1DataOut <= c1DataInDly(0) & c1DataInDly(1); c1DataValidOut <= '1'; -- Data valid asserted on phase p2. elsif c1DataValidInDly(3 downto 2) = "11" then c1DataOut <= c1DataInDly(2) & c1DataInDly(3); c1DataValidOut <= '1'; end if; end if; end process; end RTL;